You are on page 1of 25

Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Contents lists available at ScienceDirect

Engineering Applications of Artificial Intelligence


journal homepage: www.elsevier.com/locate/engappai

Sustainable building material selection: A QFD- and ELECTRE III-embedded


hybrid MCGDM approach with consensus building✩
Zhen-Song Chen a , Luis Martínez b ,∗, Jian-Peng Chang c , Xian-Jia Wang d , Sheng-Hua Xionge e ,
Kwai-Sang Chin f
a
The Key Laboratory of Safety for Geotechnical and Structural Engineering of Hubei Province, School of Civil Engineering, Wuhan
University, Wuhan 430072, China
b Department of Computer Science, University of Jaén, 23071, Jaén, Spain
c School of Business Planning, Chongqing Technology and Business School, Chongqing 400067, China
d
School of Economics and Management, Wuhan University, Wuhan 430072, China
e
College of Civil Aviation Safety Engineering, Civil Aviation Flight University of China, Guanghan 618307, China
f
Department of Systems Engineering and Engineering Management, City University of Hong Kong, 83 Tat Chee Avenue, Kowloon Tong, Hong Kong, China

ARTICLE INFO ABSTRACT


Keywords: Industrialization and urbanization have led to the rapid development of the building industry. To alleviate the
Quality function deployment adverse impact on the environment, natural resources, health and comfort of inhabitants, sustainable principles
Basic uncertain linguistic information have been incorporated into the design, construction, decoration, operation, and maintenance of buildings.
Sustainable building material selection
Among all these endeavors, sustainable building material selection is diffusely regarded as the simplest and
ELECTRE III
most essential way of achieving sustainability. To this end, this paper develops a novel hybrid multi-criteria
Multiple criteria group decision making
group decision making model for sustainable building material selection under uncertainty. The information
representation construction used for alternative-criteria assessments in this model is a brand-new notion called
Basic Uncertain Linguistic Information, which is capable of handling the qualitative performance estimates as
well as its partial reliability. Multiple stakeholders are usually responsible for sustainable building material
selection, and therefore, the fact that multiple perspectives are involved in the collective decision-making
efforts is revealed by a Consensus Reaching Process to reach a decision reflecting at least a majority-based
agreement among them. The Quality Function Deployment -based approach to determining comprehensive
importance ratings of assessment criteria is further proposed to take into account diversifying sustainable
building material requirements from different stakeholders. Subsequently, the ELECTRE III methodology, which
models the heterogeneity and non-compensation among assessment criteria as well as their qualitative essence,
is introduced to determine the ultimate priorities of alternative sustainable building materials. Finally, sensitive
and comparative analyses are performed to verify the advantages of the proposed model in comparison with
several existing methodologies.

1. Introduction in large construction practices. The stakeholders have reached a con-


sensus on the introduction of sustainable principles throughout the
Rapid industrialization and urbanization are boosting economic building life-cycle to develop a sustainable building mode which can
growth and, thereby, the rapid development of the building industry achieve a trade-off among economic, environmental and social perfor-
is the largest among developing countries, particularly China (Tian mance (Akadiri et al., 2012; Berardi, 2013; Bunz et al., 2006; Häkkinen
et al., 2018). However, growth in building construction is contributing and Belloni, 2011; Melchert, 2007). Among these, sustainable building
significantly to environment pollution and consumption of energy and
design has had the most substantial impact due to its functionality in
natural resources, thus directly influencing the health and comfort of
enhancing sustainable and technical performance and in reducing the
building occupants (Franzoni, 2011; Melchert, 2007). Reducing the
overall impact of accelerated building has been a long-term concern life-cycle cost of building (Bragança et al., 2014; Invidiata et al., 2018).

✩ No author associated with this paper has disclosed any potential or pertinent conflicts which may be perceived to have impending conflict with this work.
For full disclosure statements refer to https://doi.org/10.1016/j.engappai.2019.08.006.
∗ Corresponding author.
E-mail addresses: zschen@whu.edu.cn (Z.-S. Chen), martin@ujaen.es (L. Martínez), jpchang@ctbu.edu.cn (J.-P. Chang), wangxj@whu.edu.cn (X.-J. Wang),
xsh@my.swjtu.edu.cn (S.-H. Xionge).

https://doi.org/10.1016/j.engappai.2019.08.006
Received 26 December 2018; Received in revised form 20 June 2019; Accepted 10 August 2019
Available online 22 August 2019
0952-1976/© 2019 Elsevier Ltd. All rights reserved.
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Within the sustainable design stage, sustainable building material se- 2005). Satisfying client requirements (CRs) is therefore of extreme
lection (SBMS) presents a relaxed, yet essential way of incorporating importance because CRs define the quality of the building and,
sustainable principles into building design and its realization (Akadiri consequently, achieving CRs defines the success of a building
et al., 2013). The significance of SBMS lies in the fact that an improper project. Thus, an attempt should be made in the process of SBMS
selection of material could negatively impact the comfort and health of to take CRs into account in addition to the necessary technical
occupants and the life-cycle cost of the building. criteria of SBM.
SBMS relies on the comprehensive assessment of sustainable build- • The identification and prioritization of ACs play a fundamental
ing materials (SBMs), which commonly simultaneously factors in the role in incorporating sustainable principles in SBMS. However,
technical, economic, environmental and social performances of build- the criteria usually involve economic, environmental, social, tech-
ing materials while protecting public health over their whole life nical and aesthetical aspects. The diversity of criteria leads to the
cycle (Ortiz et al., 2009). Until now, the assessment and selection of following features: the qualitative scales of criteria satisfaction
SBM have drawn increasing attention from researchers, and a great and the non-compensation among criteria, which increase the
deal of effort has been invested in the development of appropriate complexity of the identification and prioritization process and
assessment or decision methods for SBMS. Most of the existing studies necessitate an effective MCGDM method to handle these feature
are inclined to the use of Life-Cycle Assessment (LCA)-based methods to requests.
conduct SBMS (Abeysundara et al., 2009; Arrigoni et al., 2017; Bribián • The successful implementation of SBMS needs the input of a large
et al., 2011; Marzouk et al., 2018; Najjar et al., 2017; Takano et al., amount of assessment information. Because of the inadequacy
2014, 2015). The LCA-based methods mainly focus on the assessment of of technical information of material specifications and the fuzzi-
environmental performance associated with full life cycle of a product ness of human cognition, the assessment information provided
from raw material extraction, manufacture, distribution, use, repair by stakeholders is mostly uncertain and ambiguous, hence the
and maintenance, and disposal or recycling (Cabeza et al., 2014). introduction of practical tools for characterizing and processing
However, these methodological developments fail largely in gauging uncertain information is necessary.
the economic, social and technical performances of building materials
as a whole (Akadiri et al., 2013). Besides, the processes underlying The previous analysis brings us to the investigated SBMS that is
LCA-based methods are, in many ways, costly and dependent on the a typical MCGDM problem with several specified features. Several
specific application scenarios. Breaking these limitations or drawbacks attempts have already been made to develop various MCGDM models of
forms the main motivation for us to investigate the potential of a novel structuring this type of decision making by characterizing one or more
methodology to advance the recent development of SBMS research. features as mentioned above. However, the previous MCGDM models
Essentially, SBMS is a complicated Multi-Criteria Group Decision Making proposed in literature failed to cope effectively with all these complex-
(MCGDM) process1 that characterizes an uncertain decision-making ities embedded in SBMS simultaneously, namely the incorporation of
scenario where groups of stakeholders are concerned with the appropri- CRs, the consideration of input heterogeneity, the non-compensation
ate selection of building materials; meeting not only their own diverse among criteria, and reaching consensus among the involved stakehold-
set of requirements but also objectively fulfilling several assessment ers. Even when a single type of complexity is concerned, the exist-
criteria (ACs), which usually come into conflict with each other. Given ing methods may fail to handle it effectively. For example, available
the clear-cut problem definition of SBMS, the investigated approach in models have used crisp numbers, triangular fuzzy numbers or neutro-
this paper aims to address the uncertain decision-making contexts that sophic sets to characterize assessment information (Bunz et al., 2006;
can be delineated by the following features. Govindan and Jepsen, 2016; Khoshnava et al., 2018; Mousavi-Nasab
and Sotoudeh-Anvari, 2018a,b), but are not capable of precisely and
• Building materials are the materials used for construction and flexibly modeling the basic uncertain information, which, compared
decoration of the building and encompass a wide range of types to the previous information representation formats, is better suited
such as structural materials, thermal insulation materials, and so to the qualitative need in the enhancement of uncertain assessment
on. With the development of material technology and science as descriptions.
well as the increasing demand for aesthetics and personalization, This paper aims to develop a general decision support framework
a broad range of building materials are still available for specific to help researchers and practitioners to deepen their understanding
use in almost every targeted market. For instance, the alternatives of the issue of SBMS under uncertainty by putting forward a hybrid
for flooring include rubber flooring tiles, luxury vinyl planks, MCGDM model that can deal with the complexities as mentioned
carpet tiles, solid hardwood flooring, and many more. earlier. In doing so, the model integrates five distinct components: (i)
• SBMS is a complex decision-making process involving the needs the use of Basic Uncertain Linguistic Information (BULI) to characterize
of a range of stakeholders such as clients, designers, contractors, and process uncertain assessment information, (ii) the development of
architects, and engineers. Thus, it entails group decision-making a Consensus Reaching Process (CRP) under BULI algorithm to achieve
(GDM) in which multiple individuals collectively analyze the satisfied agreement among stakeholders during collective decision mak-
issues under discussion, establish ACs, conduct assessment and ing, (iii) the incorporation of CRs into SBMS using the systematical
make the final choice from a range of alternative building mate- tool – Quality Function Deployment (QFD), (iv) the introduction of
rials. The introduction of GDM usually leads to a better decision ELimination and Choice Expressing Reality III (ELECTRE III) to determine
being made, but it gives rise to consensus problems owing to the the ultimate priorities of alternative SBMs, and, last but not least, (v) an
inherent distinction in the interests, knowledge, and experiences empirical application of the proposed model in office flooring selection
of the involved stakeholders. illustrated to verify its feasibility and rationality. These components
• During the design, construction, and decoration of a building, will be scattered throughout our descriptions of the proposed decision
clients including developers, customers, and end-users and their support framework that includes the following three stages: the struc-
requirements play basic and foundational roles (Dikmen et al., turing stage, the stage of determining ACs and their comprehensive
importance ratings (CIRs), and the assessment and exploitation stage.
1
The Multi-Criteria Decision Making (MCDM) can be viewed as a special
The general idea of the proposed hybrid MCGDM model for SBMS is
case of MCGDM where only single stakeholder is involved, and therefore, depicted in Fig. 1.
we do not distinguish MCDM particularly from MCGDM for consistency and The remainder of the paper is organized as follows. Section 2
unification purposes unless the emphasis on the terminology distinction is not presents a review of the literature on SBMS, QFD, and MCGDM meth-
only necessary but also required. ods. Section 3 defines BULI together with its comparison laws, distance

784
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Fig. 1. The general idea of the proposed hybrid MCGDM model for SBMS.

and similarity measures and aggregation operator. Section 4 provides designers, contractors, architects, engineers. Therefore, unsurprisingly,
the decision support framework and the hybrid MCGDM model for the sustainable building sector is increasingly attracting the attention
SBMS. A case study on sustainable office flooring selection is introduced of scholars and practitioners thanks to the development of novel and
in Section 5 to assess the performance of the proposed hybrid MCGDM meaningful research which is continually emerging. Berardi (2013)
model. Section 6 concludes this paper. defined sustainable building as a facility that should increase the
demand for safety and economic value, environmental responsibility
2. Literature review and resource conservation, the respect for occupant satisfaction and
stakeholder rights, and aesthetic improvements. Häkkinen and Belloni
Following the recent rapid development of economic and urban- (2011) conducted a review of related literature to specify the barriers
ization around the world, the increasing pressure from global envi- and drivers of sustainable building by carrying out a structured inter-
ronmental crises, the rapid growth of human population, depletion view and case studies in Finland in a bid to develop effective actions
of natural resources, damage to ecosystems and the constant pursuit to remove the barriers. His paper pointed out that the efficient use of
of happiness promote the creation of the concept of sustainable de- all necessary information and the active cooperation of all stakeholders
velopment in the 1987 Brundtland Report (WCED, 1987). Sustainable call for methods enabling the capture and management of the knowl-
development has drawn enormous attention from practitioners and edge. Melchert (2007) reviewed the evolution of the Dutch sustainable
researchers committed to the incorporation of sustainable principles building policy that provided essential lessons to developing countries
in a broader range of sectors of economic and social development in- according to their own experience in Holland. Bunz et al. (2006)
cluding sustainable agriculture, sustainable manufacturing, sustainable contrasted sustainable design guidelines and practices based on the life
building, sustainable transport, sustainable design. It is worth noting cycle of a building in North America, Europe, and Asia. Their paper
that the buildings in which we live, work, and play every day protect us can serve as a reference for practitioners interested in implementing
from external interference and therefore have a significant impact on sustainable design, measurement, and practices in various regions of
the health and comfort of humans. Besides, buildings are responsible the world.
for a considerable proportion of energy, electricity, water and material Any building project is rooted in the design stage, during which
consumption. In the United States, buildings account for 39% of total building materials are examined since they directly impact the sustain-
energy use, 12% of the total water consumption, 68% of the total elec- able performance throughout the building life-cycle (Franzoni, 2011).
tricity consumption, and 38% of the carbon dioxide emissions (United Building materials are the materials used for the construction and
States Environmental Protection Agency, 2009). Sustainable buildings, decoration of buildings. They include structural materials, thermal
which attempt to minimize the negative impact on human health and insulation materials, waterproof materials, sound absorption materials,
the natural environment, play a crucial role in the achievement of decoration materials, and so forth. The term SBMs, also called green
the global goal ‘‘Sustainable Cities and Communities’’, which is one building materials, refers to building materials that are environmen-
of the 17 Sustainable Development Goals (SDGs) (UN-United Nations, tally responsible, resource-efficient and not hazardous to human health
2015). SDGs emphasize the incorporation of sustainable principles into throughout the life-cycle of the buildings concerned (Franzoni, 2011).
the building life-cycle: building design, construction, decoration, oper- As described above, considerable attention has been paid by researchers
ation and maintenance and require close collaboration among clients, to the assessment and selection of SBM and LCA-based methods, which

785
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

are the most commonly used methods. Castro-Lacouture et al. (2009) the objective performance information of alternatives on each criteria,
proposed a mixed integer optimization model incorporating design such as the entropy-based method, the standard deviation method and
and budget constraints under the Leader in Energy and Environmental the maximizing deviation method; and the integrated methods on the
Design (LEED) rating system to assist in the selection of building basis of the combination of subjective methods and objective meth-
materials. Bribián et al. (2011) conducted a comparative analysis of the ods (Dong et al., 2018; Hatefi, 2019). However, the weights derived
energy and environmental impacts on commonly used building materi- by these methods cannot capture customer requirements. Up till now,
als and eco-materials based on LCA. Takano et al. (2014) and Takano the client demands for building materials have been moving towards
et al. (2015) employed LCA to study the influence of building material personalization and diversification. The capture and delivery of client
selection on environmental and economic aspects of a building in a value through satisfaction of CRs play a fundamental role in business
Finnish context, and on the life-cycle primary energy balance of a markets and directly influences the benefit and competitiveness of an
building in the Finnish context. Abeysundara et al. (2009) applied enterprise (Moghimi et al., 2017). Therefore, CRs for building materials
LCA to select sustainable materials for buildings in Sri Lanka from the should be incorporated into SBMS, which necessitates the utilization of
perspective of environmental, economic and social performance. Najjar tools and techniques for transforming CRs into the criteria which guide
et al. (2017) integrated Building Information Modeling (BIM) with SBMS.
LCA to evaluate the environmental performance of building materials QFD is an effectively customer-driven tool developed in Japan
in the construction industry. Arrigoni et al. (2017) conducted LCA to help specify the CRs and integrate the CRs into the product de-
on the environmental performance of a hempcrete-based construction sign (Akao and Mazur, 2003; Chan and Wu, 2002). QFD seeks to
material. Marzouk et al. (2018) developed a framework integrating deploy limited resources to maximize customer satisfaction. During
Life-Cycle Cost (LCC) and Leadership in Energy and Environmental De- the process of QFD, the house of quality (HOQ), which builds the
sign (LEED) to evaluate the economic and environmental performance relationship between the CRs and design requirements (DRs) of the
of building material. product, plays a fundamental and strategic role (Jin et al., 2015). The
Keeping in mind the inherent deficiencies of LCA including the process of the development of HOQ includes identifying CRs, specifying
restriction to environmental assessment, high cost, and scenario depen- importance ratings (IRs) of CRs, identifying design requirements (DRs)
dence, some researchers and scholars have turned to the adoption of which are associated with CRs, constructing the relationship between
MCDM for SBMS. Akadiri and Olomolaiye (2012) and Akadiri et al. CRs and DRs, determining the initial and modified IRs of DRs, and
(2013) identified a series of criteria related to technical, environmental determining the comprehensive importance ratings (CIRs) of DRs (Yan
and social–economic aspects on the basis of sustainable triple bottom and Ma, 2015). QFD has been applied in many fields. For instance, Dik-
men et al. (2005) conducted a case study on the building industry
line, requirements of building stakeholders, and literature reviews,
to examine the applicability of QFD as a strategic tool to facilitate
and further proposed Fuzzy Extended Analytical Hierarchy Process
marketing decisions in the construction industry and further pointed
(FEAHP) to prioritize the criteria and alternative SBMs. Sabapathy
out the limitations, breakthroughs and critical success factors of the
and Maithel (2013) developed a composite environmental index us-
applicability of QFD. Wood et al. (2016) identified end-user preferences
ing multi-criteria decision analysis and conducted a comprehensive
that affect green hospital design and incorporated them into the design
assessment of walling materials in India. Govindan et al. (2016) used
principles using QFD. Moghimi et al. (2017) put forward a hybrid
a combination of Decision-Making Trial and Evaluation Laboratory
model by combining the Means-End Chain and QFD to incorporate user
(DEMATEL) and Analytic Network Process (ANP) to analyze the interre-
values into housing design. Yuen (2014) developed a hybrid fuzzy QFD
lationships among criteria and to determine the corresponding weight
framework using cognitive network process and aggregative grading
and then applied the TOPSIS method to determine the best sustainable
clustering, and then apply it to cloud software product development.
construction materials. Khoshnava et al. (2018) put forward a hybrid
Inspired by the idea of QFD, this paper proposes a model that starts
MCDM methodology for the ranking of green building material criteria,
with the identification of CRs on the building project and materials and
in which DEMATEL was used to analyze the interrelationship between
then invokes the CRs to determine ACs and their IRs, i.e., their weights.
criteria and Fuzzy ANP (FANP) and employed it for prioritizing the
The IRs of ACs are modified by considering basic standards or needs
criteria used. Zavadskas et al. (2017) developed a hybrid MCDM model
of quality, function, and performance for the specific type of building
for residential house construction material and element selection, in
materials to determine the CIRs of the ACs.
which the neutrosophic sets were introduced to handle the uncertainty.
Further, the SWARA (Step-wise Weight Assessment Ratio Analysis) 2.2. MCGDM under uncertain context
approach was applied to determine the relative significance of the
criteria, while MULTIMOORA was used to determine the most suit- MCGDM is an extension of the MCDM method with multiple Deci-
able building materials. Tian et al. (2018) developed a hybrid MCDM sion Makers (DMs), that is used to deal with the decision problem by
method to conduct the selection of green decoration materials, in which evaluating multiple and conflicting criteria (Kabak and Ervural, 2017).
analytic hierarchy process(AHP) was used to weight the criteria. A In the MCGDM process, multiple DMs with different knowledge, expe-
gray-correlation based TOPSIS method is employed to determine the riences, and interests, collectively develop, analyze, assess alternatives
final ranking of alternatives. and finally select or determine the best one with the best interest of
The above literature review indicates that the existing literature has the whole (Nassar et al., 2003; Xiong et al., 2018). The main stages
not incorporated or translated CRs into SBMS to achieve the satisfaction of MCGDM consists of the identification of the goal, the determination
of SBM. It has also not factored in multiple points of view. Besides, the of alternatives and the committee of DMs, the specification of ACs and
human judgment models could not characterize assessment information the corresponding weights, an elicitation of the assessment information
precisely and flexibly. These gaps highlight and suggest the need for the on the performance of each alternative with respect to each AC by each
development of a hybrid MCGDM model under uncertain conditions DM, the achievement of a satisfied agreement level among DMs through
that can simulate realistic SBMS scenarios. an algorithm for CRP, the aggregation of individual opinions into
collective assessment information, and eventually the exploitation of
2.1. QFD collective assessment information using MCGDM methods to attain the
final decision. In uncertain situations, the critical issues encountered
In the existing literatures, the methods for assigning weights to during MCGDM process include the model for uncertain assessment
ACs can be classified into three categories: the subjective methods information representation formats, the algorithm for CRP under uncer-
depending on the preference information of DMs on criteria, such tain context, and the MCGDM method for exploitation under uncertain
as AHP and ANP; the objective methods developed on the basis of context.

786
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

2.2.1. CRP for GDM under uncertain context • Utility theory-based MCDM methods which try to assign a utility
In general, SBMS is an MCGDM process in which all the needs, value to each alternative, such as AHP, ANP, SAW, TOPSIS,
opinions, and concerns of the stakeholders are considered to achieve VIKOR, COPRAS, and MULTIMOORA.2
a conclusion that everyone actively involved supports, or at least can • Non-classical MCDM methods incorporating formal modeling of
accept. For instance, the clients can provide useful perspectives from a risks and uncertainty into classic MCDMs, e.g., outranking-based
user point of view, the designers can offer valuable suggestions from methods or utility theory-based methods within a fuzzy context.
an aesthetic point of view, and architects and engineers can furnish • Multi-Objective Decision Making (MODM) characterized by mul-
helpful ideas about the requirements for physical structure perfor- tiple and conflicting objective functions that are to be optimized
mances. Thus, consensus reaching among stakeholders is a fundamental under predefined constraints and are suitable for the evaluation
issue in MCGDM problems, and the achievement of an acceptable of continuous alternatives.
consensus level among individuals is preferred ahead of the aggregation
and exploitation processes. In a real SBMS process, unanimous or full Given a specific decision-making problem, the selection of the
consensus on the final outcome is difficult to attain in the wake of MCGDM method is determined in accordance with the problem defini-
the heterogeneity among the involved stakeholders, which necessitates tion and data types. As for SBMS, three features including the existence
the introduction of the notion of soft consensus (Kacprzyk, 1986) to of qualitative scales of criteria, the heterogeneity of criteria scales,
guide and control the CRP. Soft consensus, which has been developed and non-compensation among criteria have to be taken into account.
from the fuzzy majority and can be measured by distance or similarity The first feature is easy to understand. In terms of the second feature,
measures between individual opinions, releases one from strict restric- the heterogeneous scales cannot be processed directly so they need
tions on the consensus state (absence or full agreement) and therefore to be normalized a priori using appropriate normalization techniques
can reflect human perceptions of the essence of consensus (Kacprzyk, to obtain a standard measurement scale for all the criteria. However,
1986). Moreover, the CRP on soft consensus, aimed at arriving at the inevitably, this normalization process leads to a loss in the original deci-
final ranking or selection with a satisfied agreement level among DMs, sion information. The non-compensation among criteria when selecting
is a dynamic and iterative process which simulates the discussion or building materials can be illustrated by the following statement. If the
negotiation process among stakeholders, and has been widely studied emission degree of formaldehyde for alternative 𝐴 is much worse than
in the literature (Labella et al., 2018; Palomares et al., 2013, 2014a,b; that of alternative 𝐵, then alternative 𝐴 is deemed as not better than
Wu and Xu, 2016b; Rodríguez et al., 2018; Singh and Benyoucef, 2013). alternative 𝐵 even if the performances of alternative 𝐴 on other criteria
The CRP on soft consensus consists mainly of the following stages: are much better than those of alternative 𝐵.
Existing methods for SBMS have been developed using utility-based
• Consensus measures: After the elicitation of assessment infor- MCDM methods such as AHP, ANP, MULTIMOORA, TOPSIS to deter-
mation in the form of a certain representation format, the global mine the overall collective performance of the alternatives by aggre-
consensus degree need to be determined; gating all the performances of each criterion integrating the criteria
• Consensus control: With the introduction of consensus threshold weights. Utility functions are developed on the basis of the hypothesis
value, we can judge whether a predetermined global consensus that the lower performance of one alternative on a certain criterion can
degree has been reached. If the global consensus degree is sat- be compensated by attaining better performance values on other crite-
isfied, the consensus reaching process ends, and the aggregation ria. Therefore, these methods are unable to cope with heterogeneity
and exploitation processes are activated. Otherwise, a new discus- and non-compensation among the criteria.
sion or negotiation is performed with the assistance of a feedback Meanwhile, ELECTRE methods have drawn a lot of attention from
mechanism used to provide stakeholders with suggestions for researchers and practitioners in view of a series of intriguing mer-
improving the global consensus level. its in handling the coexistence of qualitative and quantitative scales,
• Feedback mechanism: Based on the feedback information, DMs heterogeneity and non-compensation (Chavira et al., 2017; Figueira
with the least contributions to a sufficiently high consensus level et al., 2013; Govindan and Jepsen, 2016; Sánchez-Lozano et al., 2016;
adjust their assessment information to reach a higher consensus Vasto-Terrientes et al., 2015; Ferreira et al., 2016; Yuen, 2009a).
level. After the adjustment, the first iteration is ended and the ELECTRE-based methods were developed on the basis of a certain
second iteration is activated until the consensus threshold value attractive relationship built on the strength of the evidence supporting
is achieved or a maximum number of iterations is reached. the assertion that ‘‘a is at least as good as b’’ (concordance) and the
strength of the evidence against this assertion (discordance). Given that
Given that the consensus is related to GDM problems, CRP can be the aim of SBMS is to determine the best suitable building material
introduced into MCGDM, QFD or other methods with multiple individ- for fulfilling specific requirements, we can introduce ELECTRE III into
uals involved. As the uncertain assessment information is represented SBMs. One more reason is that, compared with other family members,
in the form of BULI, we put forward an algorithm for CRP under BULI ELECTRE III has a more important advantage concerning copying with
context to facilitate the achievement of a satisfactory consensus level. inaccuracy, imprecision or uncertainty of DM’s preference information
through the introduction of pseudo-criteria. The criteria have been
2.2.2. MCGDM methods constructed by the indifference, preference or veto thresholds. The
MCGDM aims at arriving at a reasonable decision considering mul- objective of this paper is to develop a new version of ELECTRE III
tiple conflicting criteria and has become an relatively independent in which the value is represented in the form of BULI. Two main
branch of operations research. In fact, MCGDM methods can be ex- steps are followed.First, binary outranking relations for each pair of
tended from the MCDM theory and its methodologies, and a consid- alternatives are constructed on the basis of the partial concordance
erable number of MCDM methods have been developed and applied to and discordance within the context of BULI. Second, the outranking
many areas including sciences, business, production, and engineering. relations are exploited via distillation methods to obtain a partial
Basically, the existing MCDM methods can be classified into the fol- pre-order of alternatives.
lowing four categories (Greco et al., 2016; Kumar et al., 2017; Yuen,
2009b; Chen et al., 2019): 2
Details of these acronyms are: simple additive weighting(SAW), ana-
lytic hierarchy process(AHP), techniques for order preference by similarity
• Outranking-based MCDM methods that depend on outranking to an ideal solution(TOPSIS), Vlsekriterijumska Optimizacija IKompromisno
relation which is the binary relation between pairs of alterna- Resenje(VIKOR), complex proportional assessment(COPRAS), Multiple Objec-
tives, e.g., the family of ELECTRE methods and the family of tive Optimization on the basis of Ratio Analysis plus Full Multiplicative
PROMETHEE methods. Form(MULTIMOORA).

787
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

{
3. Basic uncertain linguistic information 𝑆 𝑅 = 𝑠𝑅
0
= very low, 𝑠𝑅
1
= low, 𝑠𝑅
2
= slightly low, 𝑠𝑅
3
= moderate,
𝑠𝑅
4
= slightly high, 𝑠𝑅
5
= high,
}
In a complex and uncertain decision context, DMs prefer using qual- 𝑠𝑅 = very high ,
{ 6
itative linguistic representations rather than qualitative information to 𝑆 𝑃 = 𝑠𝑃0 = very poor, 𝑠𝑃1 = poor, 𝑠𝑃2 = slightly poor, 𝑠𝑃3 = medium,
articulate their judgements (Chen et al., 2018a). The reason behind 𝑠𝑃4 = slightly good, 𝑠𝑃5 = good,
this is that the qualitative linguistic representations are closer to the }
𝑠𝑃6 = very good .
cognitive process of human beings and provide more flexibility and
exert less burden for DMs to express their assessments over certain Several linguistic computational models have been developed over
objects. The fuzzy linguistic approach, which has been developed using the past few decades to facilitate the process of linguistic informa-
the fuzzy sets theory as its main theoretical foundation, is used to tion. The most renowned is the 2-tuple linguistic representation model
proposed by Herrera and Martínez (2000), which has been consen-
model linguistic information by using the concept of linguistic variables
sually recognized as an efficient computational approach to show-
that approximate to humans’ cognitive process (Rodríguez et al., 2016;
ing the discrete linguistic terms in a continuous way. The linguis-
Zadeh, 1975). Since its inception, the fuzzy linguistic approach has
tic computational manipulation based on the 2-tuple linguistic repre-
drawn dramatically increasing attention from researchers and prac- sentation model demonstrates the advantages of the convenience in
titioners in consideration of the outstanding advantages of modeling improving interpretability and the avoidance of the loss of informa-
uncertainty in complex decision-making situations (Rodríguez et al., tion (Rodríguez et al., 2016). Mathematically, the 2-tuple linguistic
2016). The recent developments of this approach provide DMs more representation model successfully extends the traditional discretely
freedom in utilizing richer linguistic information to express their judg- distributed LTS to a continuous version, and the basic knowledge
ments and opinions. To address the gap, the attempts have been made relating to it is briefly reviewed below citing from Herrera and Martínez
to extend and improve the fuzzy linguistic approach from various (2000), Martínez et al. (2010, 2015) and Martínez and Herrera (2012)
perspectives, such as unbalanced linguistic term sets (Herrera et al., for the benefit of any follow-up studies.
2008), hesitant fuzzy linguistic term sets (HFLTSs) (Rodríguez et al., { }
2012), proportional HFLTS (Chen et al., 2016), and HFLTS possibility Definition 1. Let 𝑆 = 𝑠0 , 𝑠1 , … , 𝑠𝜏 be an LTS and 𝑆̄ the 2-tuple set
associated with 𝑆 defined as 𝑆̄ = 𝑆 × [−0.5, 0.5). Let 𝜓 ∈ [0, 𝜏] be a
distribution (Wu and Xu, 2016a). Among these extensions, HFLTS pos-
value representing the result of a symbolic aggregation operation. The
sibility distribution is receiving significant attention from researchers
2-tuple that expresses the equivalent information to 𝜓 is then obtained
by reason of its addition of the extra dimension of possibility degrees as:
of linguistic terms to enhance the quality of assessment representa-
tion (Chen et al., 2017, 2018b). It is noteworthy that the prevalent 𝛥𝑆 ∶ [0, 𝜏] → 𝑆 × [−0.5, 0.5) ,
{
fuzzy linguistic approaches have not considered the reliability degree ( ) 𝑠𝑖 , 𝑖 = round (𝜓)
where 𝛥𝑆 (𝜓) = 𝑠𝑖 , 𝛼 , with .
originating in part from the high possibility in information processing 𝛼 = 𝜓 − 𝑖, 𝛼 ∈ [−0, 5, 0.5)
of suffering from a partial lack of certainty, mostly because of rounding
{ } ( )
operations, indistinguishability of the measurement tools or subjec- Proposition 1. Let 𝑆 = 𝑠0 , 𝑠1 , … , 𝑠𝜏 be an LTS and 𝑠𝑖 , 𝛼 ∈ 𝑆̄ be a
tive confidence when introducing data, etc.(Chen et al., 2018a; Jin −1 ̄
linguistic 2-tuple value. The inverse function, 𝛥𝑆 ∶ 𝑆 → [0, 𝜏], of the one
et al., 2018; Mesiar et al., 2018). The inclusion of reliability dimension to one mapping 𝛥𝑆 is defined as:
into the linguistic evaluation would facilitate to address these issues. ( )
𝛥−1
𝑆 𝑠𝑖 , 𝛼 = 𝑖 + 𝛼.
Inspired by Basic Uncertain Information (BUI) which is a two-tuple ( )
containing the crisp number and its reliability degree lying in the unit Therefore, a linguistic 2-tuple value 𝑠𝑖 , 𝛼 ∈ 𝑆̄ can be identified
interval (Jin et al., 2018; Mesiar et al., 2018), we propose a brand-new with a numerical value in the interval of granularity of 𝑆, i.e., [0, 𝜏].
approach called BULI for modeling and processing uncertain assessment Once the linguistic 2-tuple model is utilized to accomplish the processes
of CW in linguistic decision making, the ordering of linguistic 2-tuple
information and define elaborated computational manipulations for it,
values is usually required and can be carried out by the following
which include comparison laws, distance and similarity measures, and
ordinary lexicographic order.
aggregation functions.
The implementation of a fuzzy linguistic approach necessitates the ( ) ( )
Definition 2. Let 𝑠𝑘 , 𝛼 and 𝑠𝑙 , 𝛾 be two linguistic 2-tuples. Then:
selection of an appropriate LTS and corresponding semantics, which im- ( ) ( )
(i) if 𝑘 < 𝑙, then 𝑠𝑘 , 𝛼 is smaller than 𝑠𝑙 , 𝛾 ;
plies the necessity of a computing with words (CW) process (Martínez (ii) if 𝑘 = 𝑙, then
et al., 2010; Torres et al., 2018). LTS is constructed as an ordered ( ) ( )
(a) if 𝛼 = 𝛾, then 𝑠𝑘 , 𝛼 and 𝑠𝑙 , 𝛾 represent the same information;
structure in which the linguistic terms involved are usually distributed ( ) ( )
(b) if 𝛼 < 𝛾, then 𝑠𝑘 , 𝛼 is smaller than 𝑠𝑙 , 𝛾 .
uniformly within a predefined domain. Generally, the semantics of the
linguistic terms in an LTS are specified in accordance with the specific { } ( )
Definition 3. Let 𝑆 = 𝑠0 , 𝑠1 , … , 𝑠𝜏 be an LTS and 𝑠𝑖 , 𝛼 ∈ 𝑆̄ be a
decision scenario involved. Without loss of generality, we assume that
{ } linguistic 2-tuple value. The negation operator for a linguistic 2-tuple
a LTS can be denoted by 𝑆 = 𝑠0 , 𝑠1 , … , 𝑠𝜏 with the granularity being is defined as:
𝜏 + 1. In the context of SBMS, in order to facilitate the articulation of (( )) ( ( ))
the preference information of CRs or ACs, the LTSs 𝑆 𝐼 , 𝑆 𝑅 , and 𝑆 𝑃 𝑁𝑒𝑔 𝑠𝑖 , 𝛼 = 𝛥𝑆 𝜓 − 𝛥−1 𝑆 𝑠𝑖 , 𝛼 .
are predefined for the measurement of the relationships between CRs
The 2-tuple linguistic representation model has appeared in thou-
and ACs, the decision team, and the weights of the performances of
sands of scholarly examined papers and monographs. More details can
alternatives under various criteria, respectively. The LTSs 𝑆 𝐼 , 𝑆 𝑅 and
be found in recently published papers and monographs (Herrera and
𝑆 𝑃 are detailed as follows.
{ Martínez, 2000; Martínez et al., 2010, 2015; Martínez and Herrera,
𝑆 𝐼 = 𝑠𝐼0 = very unimportant, 𝑠𝐼1 = unimportant, 𝑠𝐼2 = slightly unimportant, 2012).
𝑠𝐼3 = medium, BUI proposed by Jin et al. (2018) and later utilized in Mesiar
}
𝑠𝐼4 = slightly important, 𝑠𝐼5 = important, 𝑠𝐼6 = very important , et al. (2018) is characterized by a real number 𝑥 ∈ [0, 1] and its
degree of reliability 𝑐 ∈ [0, 1], which can be represented as ⟨𝑥; 𝑐⟩ ∈

788
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807
( ( ))𝑛 ( )
[0, 1] × [0, 1]. Synthesizing the merits of BUI with the 2-tuple linguistic 𝖡𝖴𝖫𝖨𝖶𝖠 ∶  𝑆 →  𝑆 and can be defined as
representation model, the concept of BULI is proposed below to extend
the capability of BUI in modeling the complex human reasoning under (⟨ ( ) ⟩ ⟨ ( ) ⟩)
𝖡𝖴𝖫𝖨𝖶𝖠 𝛥𝑆 𝜓1 ; 𝑐1 , … , 𝛥𝑆 𝜓𝑁 ; 𝑐𝑁
uncertainty.
⟨ ( ( ) ( )) (
{ } = 𝖳𝖶𝖠 𝛥𝑆 𝜓1 , … , 𝛥𝑆 𝜓𝑁 ; 1 − 𝖶𝖠 1 − 𝑐1 + 𝑐1 𝜓1 , … ,
Definition 4. Let 𝑆 = 𝑠𝛼 |𝛼 = 0, 1, … , 𝜏 be an LTS. The pair 𝑏𝑖 = ) ( )⟩
⟨ ( ) ⟩ ( )
𝛥𝑆 𝜓𝑖 ; 𝑐𝑖 , which binarily includes a 2-tuple linguistic term 𝛥𝑆 𝜓𝑖 1 − 𝑐𝑁 + 𝑐𝑁 𝜓𝑁 + 𝖶𝖠 𝑐1 𝜓1 , … , 𝑐𝑁 𝜓𝑁 (2)
defined on 𝑆̄ and its source reliability 𝑐𝑖 ∈ [0, 1], is called a representa- ⟨ ( ( ) ( )) ( )⟩
tion of Basic Uncertain Linguistic Information (BULI). = 𝖳𝖶𝖠 𝛥𝑆 𝜓1 , … , 𝛥𝑆 𝜓𝑁 ; 𝖶𝖠 𝑐1 , … , 𝑐𝑁
⟨ 𝑁 ⟩
∑ ∑
𝑁
= 𝛥( 𝑤𝑛 𝜓𝑛 ); 𝑤𝑛 𝑐 𝑛 ,
BULI is suitable for clients and other stakeholders while expressing 𝑛=1 𝑛=1
their preferences or judgements in the case of SBMS. For example, where 𝖳𝖶𝖠 is the 2-tuple weighted averaging (WA) operator and 𝖶𝖠
a DM may give a comment on the performance of an alternative is the plain WA operator.
material on a specific criterion, which is in the form of BULI and
⟨ ⟩ The following example illustrates the concept of BULI and its com-
represented as 𝑠𝑃4 ; 0.7 , in which the former indicates that the individ-
parison laws, distance measure, and aggregation function in the context
ual considers that the performance of the alternative on the criterion
of SBMS.
is ‘‘slightly good’’, and the latter represents that the reliability level
of this judgement or their related experience and knowledge is 0.7.
Example 1. For a specific SBMS problem in which three alternative
To facilitate the processing of BULI, we now present its instrumental { }
SBMs represented as 𝐵𝑀1 , 𝐵𝑀2 , 𝐵𝑀3 are available and three as-
computational manipulations such as comparison laws, distance and { }
sessment criteria denoted by 𝐴𝐶1 , 𝐴𝐶2 , 𝐴𝐶3 are underpinned for
similarity measures and aggregation operator. { }
alternative evaluation, three DMs denoted by 𝐷𝑀1 , 𝐷𝑀2 , 𝐷𝑀3 are
⟨ ( ) ⟩ ⟨ ( ) ⟩ requested to provide their judgements on each SBM with respect to
Definition 5. Let 𝑏𝑖 = 𝛥𝑆 𝜓𝑖 ; 𝑐𝑖 and 𝑏𝑗 = 𝛥𝑆 𝜓𝑗 ; 𝑐𝑗 be arbitrary
each criterion in the form of BULI based on LTS 𝑺 𝑃 . The assessment
two BULI pairs, the comparison laws between them can be defined by
( ) (( ) ( )) information is presented as follows.
(1) 𝑏𝑖 > 𝑏𝑗 ⇔ 𝜓𝑖 𝑐𝑖 > 𝜓𝑗 𝑐𝑗 ∨ 𝜓𝑖 𝑐𝑖 = 𝜓𝑗 𝑐𝑗 ∧ 𝜓𝑖 > 𝜓𝑗 ;
( ) (( ) ( )) ( ) (⟨ ( 1 ) 1 ⟩)
(2) 𝑏𝑖 < 𝑏𝑗 ⇔ 𝜓𝑖 𝑐𝑖 < 𝜓𝑗 𝑐𝑗 ∨ 𝜓𝑖 𝑐𝑖 = 𝜓𝑗 𝑐𝑗 ∧ 𝜓𝑖 < 𝜓𝑗 ;
( ) ( ) 𝑩 1 = 𝑏1𝑚𝑛 3×3 = 𝛥𝑆 𝜓𝑚𝑛 ; 𝑐𝑚𝑛 3×3
(3) 𝑏𝑖 = 𝑏𝑗 ⇔ 𝜓𝑖 = 𝜓𝑗 ∧ 𝑐𝑖 = 𝑐𝑗 . ⟨ ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩
⎛ 𝑠𝑃5 ; 0.8 𝑠6 ; 0.7 𝑠5 ; 0.9 ⎞
⎜⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩⎟
{ } ( ) = ⎜ 𝑠4 ; 0.9 𝑠5 ; 0.9 𝑠3 ; 0.9 ⎟ ,
A set of BULI pairs 𝑏𝑖 || 𝑖 = 1, 2, … , 𝑛 on  𝑆 is said to be ordered ⟨
⎜ 𝑃 ⟩ ⟨ ⟩ ⟨ ⟩⎟
⎝ 𝑠6 ; 0.6 𝑠𝑃5 ; 0.8 𝑠𝑃5 ; 0.7 ⎠
if they are arranged in decreasing order. The antisymmetry in Defini-
tion 5 eliminates the uncertain cases when (both ( ) (⟨ ( 2 ) 2 ⟩)
) 𝑏𝑖 precedes 𝑏𝑗 and 𝑏𝑗 𝑩 2 = 𝑏2𝑚𝑛 3×3 = 𝛥𝑆 𝜓𝑚𝑛 ; 𝑐𝑚𝑛 3×3
precedes 𝑏𝑖 . Meanwhile, all BULI pairs on  𝑆 are comparable under ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩
⎛ 𝑠5 ; 0.7 𝑠5 ; 1.0 𝑠5 ; 0.9 ⎞
the relation, meaning that the property of ‘‘totality’’ is satisfied. The ⎜⟨ ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩⎟
= ⎜ 𝑠𝑃6 ; 0.6 𝑠5 ; 1.0 𝑠4 ; 0.9 ⎟ ,
proposed comparison laws are implicitly implied with transitivity
( ) and ⎜⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩⎟
therefore form a strict total ordering of BULI pairs over  𝑆 . ⎝ 𝑠5 ; 0.7 𝑠4 ; 1.0 𝑠5 ; 0.9 ⎠
( ) (⟨ ( 3 ) 3 ⟩)
𝑩 3 = 𝑏3𝑚𝑛 3×3 = 𝛥𝑆 𝜓𝑚𝑛 ; 𝑐𝑚𝑛 3×3
Definition 6. Let 𝑆 be defined as before. For any two BULI pairs 𝑏𝑖 = ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩
⟨ ( ) ⟩ ⟨ ( ) ⟩ ⎛ 5𝑠 ; 0.9 𝑠 ; 0.8 𝑠5 ; 1.0 ⎞
𝛥𝑆 𝜓𝑖 ; 𝑐𝑖 and 𝑏𝑗 = 𝛥𝑆 𝜓𝑗 ; 𝑐𝑗 , the distance measure between ⎜⟨ 𝑃 ⟩ ⟨ 𝑃
5
⟩ ⟨ 𝑃 ⟩⎟
them can be defined as: = ⎜ 𝑠4 ; 0.9 𝑠6 ; 0.7 𝑠4 ; 0.9 ⎟ .
⎜⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩⎟
| | | | ⎝ 𝑠4 ; 0.9 𝑠5 ; 0.9 𝑠5 ; 0.8 ⎠
( ) |𝜓𝑖 𝑐𝑖 − 𝜓𝑗 𝑐𝑗 || + ||𝜓𝑖 − 𝜓𝑗 || ⟨ ( 1) 1⟩ ⟨ ⟩
𝑑 𝑏𝑖 , 𝑏𝑗 = | (1) Take 𝑏111 = 𝛥𝑆 𝜓11 ; 𝑐11 = 𝑠𝑃5 ; 0.8 in 𝑩 1 for example, 𝑠𝑃5
𝜏 +1
( ) indicates that 𝐷𝑀1 considers that the performance of alternative 𝐵𝑀1
where 𝑑 𝑏𝑖 , 𝑏𝑗 ∈ [0, 1]. on criterion 𝐴𝐶1 is ‘‘good’’, but he thinks that the reliability level of his
judgement is 0.8 in consideration of his own experience and knowledge.
The assessment information given by 𝐷𝑀1 on the performance of
Evidently, the distance measure satisfies the following properties:
( ) ( ) ⟨ ⟩ ⟨ ⟩
the three alternatives on criterion 𝐴𝐶1 are 𝑠𝑃5 ; 0.8 , 𝑠𝑃4 ; 0.9 and
(a) 0 ≤ 𝑑 𝑏𝑖 , 𝑏𝑗 ≤ 1, (b) 𝑑 𝑏𝑖 , 𝑏𝑗 = 0 if and only if 𝑏𝑖 = 𝑏𝑗 , and
( ) ( ) ⟨ 𝑃 ⟩
c) 𝑑 𝑏𝑖 , 𝑏𝑗 = 𝑑 𝑏𝑗 , 𝑏𝑖 . Given that similarity measure and distance 𝑠6 ; 0.6 , respectively. In accordance with Definition 5, we can easily
measure are dual concepts and the former is the compensation of the obtain 𝑏111 > 𝑏131 > 𝑏121 , which indicates that from the perspective of
( ) ( ) 𝐷𝑀1 , 𝐵𝑀1 performs better than 𝐵𝑀2 and 𝐵𝑀3 in terms of criterion
latter, we can obtain the similarity measure 𝑠𝑖𝑚 𝑏𝑖 , 𝑏𝑗 as 𝑠𝑖𝑚 𝑏𝑖 , 𝑏𝑗 =
( ) 𝐴𝐶1 . Besides, we can calculate the distance between 𝑏111 and 𝑏131 as per
1 − 𝑑 𝑏𝑖 , 𝑏𝑗 using the Zadeh negation. We acknowledge that similarity
measure has the similar properties to that of distance measure, and the Definition 6, the result of which is given by
( )
closer 𝑠𝑖𝑚 𝑏𝑖 , 𝑏𝑗 is to 1, the more similar 𝑏𝑖 and 𝑏𝑗 is. ( ) |5 × 0.8 − 6 × 0.6| + |5 − 6| 0.4 + 1
𝑑 𝑏111 , 𝑏131 = = = 0.2.
Furthermore, for the achievement of the aggregation of BULI, we 6+1 7
propose a BULI aggregation operator referring to the BUI-aggregation Given that the weights of 𝐷𝑀1 , 𝐷𝑀2 and 𝐷𝑀3 are 0.4, 0.3, and 0.3,
function proposed by Mesiar et al. (2018). respectively, the collective judgements on each alternative with respect
to each criterion can be derived in the use of the BULIWA operator as
{
⟨ ( ) ⟩| ( )
Definition 7. Let 𝑩 = 𝑏𝑛 = 𝛥𝑆 𝜓𝑛 ; 𝑐𝑛 | 𝛥𝑆 𝜓𝑛 ∈ 𝑆, 𝑐𝑛 ∈ [0, 1] , below.
| ( ) (⟨ ( ) ⟩)
𝑛 = 1, 2, … , 𝑁} be the set of BULI pairs to be aggregated, in which 𝑩 = 𝑏𝑚𝑛 3×3 = 𝛥𝑆 𝜓𝑚𝑛 ; 𝑐𝑚𝑛 3×3
⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩
𝑆 refers to a given LTS. The weighting vector is denoted by 𝑾 =
∑ ⎛ 𝑠5 , 0 ; 0.8 𝑠5 , 0.4 ; 0.82 𝑠5 , 1 ; 0.93 ⎞
(𝑤1 , … , 𝑤𝑁 ) with two fundamental conditions 𝑤𝑛 ∈ [0, 1] and 𝑁 ⎜⟨( ) ⟩ ⟨( ) ⟩ ⟨( ) ⟩⎟
𝑛=1 𝑤𝑛 = ⎜ 𝑠𝑃5 , −0.4 ; 0.81 𝑠𝑃5 , 0.3 ; 0.87 𝑠𝑃4 , −0.1 ; 0.9 ⎟ .
= 1 satisfied. Then, the elements in 𝑩 can be aggregated by the BULI ⎜ ⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩ ⎟
weighted averaging (BULIWA) operator, which is a mapping function ⎝ 𝑠5 , 1 ; 0.72 𝑠5 , −0.3 ; 0.89 𝑠5 , 1 ; 0.79 ⎠

789
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

4. Decision support framework and decision model for SBMS and determining the CIRs of ACs. In the process, the initial IRs of
ACs reflect the contribution levels of ACs to the fulfillment of CRs
In this section, we proposed the general framework of the decision while the modified IRs of ACs measure the relative contribution
support model for SBMS. An algorithm for CRP with BULI is designed level of ACs to meeting basic standards or needs of quality and
as Algorithm 1 in Section 4.2 in a bid to reach a satisfactory consensus function of this type of building material. Finally, the combination
level among the decision appraisals provided by DMs. The process of of the initial IRs of ACs and the modified IRs of ACs give the CIRs
determining the ACs and the corresponding CIRs with the assistance of ACs. Besides, given the complexity and uncertainty associated
of the QFD method is presented in Section 4.3. The last part of this with SBMS, clients CR preferences, the relationship degrees be-
section, i.e., Section 4.4, is devoted to the assessment and exploitation tween CRs and ACs and the modified IRs of ACs are represented
processes. in the form of BULI to facilitate the elicitation of the assessment
information. Finally, the algorithm for CRP under BULI is intro-
4.1. The general framework of decision support model for SBMS duced to ensure the collective assessment information is obtained
with a satisfied consensus level among stakeholders.
In real-life SBMS scenarios, the stakeholders conduct selection based • Stage 3: Assessment and exploitation. In our general decision
on their respective knowledge and experience, which often leads to support framework, this stage includes the assessment of the
poor decisions and further influences the building quality and client performance of alternatives for each criterion and the exploitation
satisfaction. In order to guide the stakeholders to accomplish SBMS of assessment information to attain the most suitable material. In
effectively, we develop a general decision support framework for SBMS the hybrid MCGDM model, DMs articulate their assessment infor-
that includes three main stages: the structuring stage, the stage of mation regarding the performances of alternatives on criteria in
determining ACs and their CIRs or weights, and the assessment and the form of BULI. CRP and aggregation processes are subsequently
exploitation stage. To address the issues arising with existing MCGDM performed to obtain collective assessment information with a
models for SBMS, we will further develop a hybrid MCGDM that works satisfactory consensus. Finally, BULI-based ELECTRE-III is applied
under uncertain conditions. In this framework, the BULI along with to exploit the collective assessment information to achieve a final
its operational laws are used to characterize and process uncertainty outranking of alternative building materials.
while the algorithm for CRP under BULI is designed to cope with
consensus problems. Further, a BULI-based QFD method is put forward
4.2. CRP under BULI
to determine ACs and their CIRs, while BULI-based ELECTRE-III is
proposed to conduct the selection process.
In this section, we develop a novel algorithm for CRP under BULI
• Stage 1: Structuring. Firstly, we define the scope of the SBMS to facilitate the achievement of satisfied consensus level during GDM.
problem including the backgrounds and goals, the stakeholders Without loss of generality, the considered CRP in SBMS involves 𝑇
{ }
and their responsibilities, the methods for characterizing and DMs that are denoted by 𝐷𝑀1 , … , 𝐷𝑀𝑡 , … , 𝐷𝑀𝑇 , 𝑀 potential al-
{ }
processing uncertainty and dealing with consensus, the methods ternatives denoted by 𝐴1 , … , 𝐴𝑚 , … , 𝐴𝑀 , and 𝑁 criteria denoted by
{ }
for determining ACs and their CIRs and for performing the se- 𝐶1 , … , 𝐶𝑛 , … , 𝐶𝑁 . The DMs articulate their assessment information
lection. The goal of the proposed MCGDM model is to select in the form of BULI pairs which can be denoted by the decision matrices
{ 𝑡 ( 𝑡 ) } ⟨ ( 𝑡 ) 𝑡 ⟩
the most appropriate SBM which provides not only basic func- 𝑩 = 𝑏𝑚𝑛 𝑀×𝑁 |𝑡 = 1, … , 𝑇 , where 𝑏𝑡𝑚𝑛 = 𝛥𝑆 𝜓𝑚𝑛 ; 𝑐𝑚𝑛 indicates
tions and requirements but also satisfies aesthetic and sustainable the performance value of alternative 𝐴𝑚 on criterion 𝐶𝑛 given by 𝐷𝑀𝑡 ,
requirements. Since SBMS is an important part of the design, and 𝑐𝑚𝑛 𝑡 ∈ [0, 1] represents the reliability degree of this assessment
construction, and decoration of the sustainable building, the de- given by 𝐷𝑀𝑡 . With all these denotations clarified we can design the
cision team responsible for SBMS mainly includes the designers, algorithm for CRP under BULI context which includes three main steps,
architects, contractors, and engineers. Besides, given the fact i.e., the consensus measure definition, the consensus control, and the
that CRs directly influence SBMS, the clients including investors, feedback mechanism, which are inspired by Wu and Xu (2016b) and
purchasers, occupants, and end-users impose a range of require- are detailed as below followed by brief algorithmic descriptions of the
ments on sustainable buildings reflecting both requirements and flow chart presented in Fig. 2.
preferences. To effectively and efficiently handle uncertainty, (1) Consensus measure.
BULI together with its computational manipulations such as com- The consensus degree is determined to measure the agreement level
parison laws, distance and similarity measures and aggregation amongst the DMs and can be calculated based on similarity measures.
( )
operators are put forward depending on BUI and certain fuzzy 𝑡 𝑡
Firstly, we can obtain a similarity matrix 𝑪𝑫 𝑡1 𝑡2 = 𝑐𝑑𝑚𝑛
1 2
for
linguistic approaches. To achieve consensus to a satisfactory de- 𝑡1 𝑡2
𝑀×𝑁
gree among stakeholders, we further design an algorithm for CRP each pair of DMs, where 𝑐𝑑𝑚𝑛 indicates the similarity measure between
under BULI. To incorporate the CRs into SBMS in Stage 2, we 𝐷𝑀𝑡1 and 𝐷𝑀𝑡2 on the performance of alternative 𝐴𝑚 with respect to
𝑡1 𝑡2
propose a BULI-based QFD method that helps determine the ACs the criterion 𝐶𝑛 , 𝑡1 , 𝑡2 = 1, … , 𝑇 , 𝑡1 ≠ 𝑡2 . The closer 𝑐𝑑𝑚𝑛 is to 1, the
and their CIRs. To avoid the negative impact of heterogeneity more similar the assessment information of 𝐴𝑚 on 𝐶𝑛 given by 𝐷𝑀𝑡1
𝑡1 𝑡2
and non-compensatory among ACs in Stage 3, we also propose and 𝐷𝑀𝑡2 will be. 𝑐𝑑𝑚𝑛 can be calculated by the following equation.
a BULI-based ELECTRE-III method that can conduct assessment | 𝑡1 𝑡1 𝑡2 𝑡2 | | 𝑡1 𝑡2 |
( ) |𝜓𝑚𝑛 𝑐𝑚𝑛 − 𝜓𝑚𝑛 𝑐𝑚𝑛 | + |𝜓𝑚𝑛 − 𝜓𝑚𝑛 |
=1− | | | |
and exploitation. 𝑡1 𝑡2 𝑡1 𝑡2
𝑐𝑑𝑚𝑛 = 𝑠𝑖𝑚 𝑏𝑚𝑛 , 𝑏𝑚𝑛 (3)
• Stage 2: Determination of ACs and their CIRs. Regarding decision 𝜏 +1
support, the goals of SBMS should be divided into smaller goals ( )
Secondly, we can attain the consensus matrix 𝑪𝑫 = 𝑐𝑑𝑚𝑛 𝑀×𝑁 by
that are easy to interpret and evaluate and can be deemed as
aggregating all the similarity matrices, in which 𝑐𝑑𝑚𝑛 indicates the
ACs. Besides, CIRs of ACs, also referred to as weights of ACs, are
consensus measure among the DMs on the performance of alternative
introduced to represent their relative influence levels on SBMS.
𝐴𝑚 with respect to the criterion 𝐶𝑛 . The closer 𝑐𝑑𝑚𝑛 is to 1, the greater
Also, QFD is introduced to translate CRs into ACs, including
the agreement among DMs on the performance of 𝐴𝑚 on 𝐶𝑛 will be.
the following steps: identifying the targeted clients and the CRs
𝑐𝑑𝑚𝑛 can be obtained by the following equation
and specifying their preferences for CRs; identifying ACs and
∑𝑇 ∑𝑇 𝑡1 𝑡2
determining the degrees of relationship between CRs and ACs; 𝑡1 =1 𝑡2 =1;𝑡2 ≠𝑡1 𝑐𝑑𝑚𝑛
determining initial IRs of ACs; specifying the modified IRs of ACs, 𝑐𝑑𝑚𝑛 = (4)
𝑇 (𝑇 − 1)∕2

790
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Fig. 2. The flow chart of the CRP with BULI inputs.

( )
With 𝑪𝑫 = 𝑐𝑑𝑚𝑛 𝑀×𝑁 we can further determine the global con- modified is determined as follows:
sensus degree 𝑐𝑑, which is used to measure the global agreement { }
( )| { }
level among the DMs and to control the CRP. 𝑐𝑑 is calculated by the 𝐴𝑚∗ , 𝐶𝑛∗ ||𝑐𝑑𝑚∗ 𝑛∗ = 𝑐𝑑 = min 𝑐𝑑𝑚𝑛 |𝑚 = 1, … , 𝑀; 𝑛 = 1, … , 𝑁 .
following equation | 𝑚,𝑛

{ } (6)
𝑐𝑑 = min 𝑐𝑑𝑚𝑛 |𝑚 = 1, … , 𝑀; 𝑛 = 1, … , 𝑁 (5)
𝑚,𝑛 Subsequently we determine the DM who makes the least con-
( )
tributions on the consensus level in the position 𝐴𝑚∗ , 𝐶𝑛∗ as
(2) Consensus control.
follows.
Consensus control is used to control the level of consensus and the { | ∑ { 𝑇 }}
number of iterations of CRP to be carried out. Firstly, we determine | 𝑇 ∑
𝐷𝑀𝑡∗ || 𝑡∗ 𝑡
𝑐𝑑 ∗ ∗ = min 𝑐𝑑𝑚∗ 𝑛∗ |𝑙 = 1, … , 𝑇
𝑙𝑡
. (7)
a minimum consensus threshold value 𝛾 ∈ [0, 1] with respect to the |𝑡=1,𝑡≠𝑡∗ 𝑚 𝑛 𝑙
| 𝑡=1,𝑡≠𝑙
specific decision scenario. Next, we determine whether the obtained
Combining Eq. (6) and Eq. (7), we can determine the position
global consensus degree is greater than the threshold value 𝛾 or not. If
where assessment information needs to be changed and the DM
𝑐𝑑 > 𝛾, CRP ends, and the aggregation and exploitation processes are
who needs to change.
activated. Otherwise, we should apply a feedback mechanism to modify
the assessment information. Furthermore, in order to avoid situations • Direction process
where consensus measure never reaches an appropriate satisfaction This process is utilized to determine the direction of the change
value, we incorporate a maximum number of iterations which are to be suggested to 𝐷𝑀𝑡∗ in an effort to reach a higher consensus
level. To achieve this, we compute the collective assessment
denoted as 𝙼𝚊𝚡𝙸𝚝𝚎. If the number of iterations is greater than 𝙼𝚊𝚡𝙸𝚝𝚎, ( )
CRP ends. Otherwise, the feedback mechanism starts. information 𝑏𝑚∗ 𝑛∗ in the position 𝐴𝑚∗ , 𝐶𝑛∗ which guides the
direction of the change. Without loss of generality, we denote the
(3) Feedback mechanism. { }
weighting vector of DMs by 𝜆1 , … , 𝜆𝑡 , … , 𝜆𝑇 with 0 ≤ 𝜆𝑡 ≤ 1
A feedback mechanism is introduced to provide suggestions to DMs ∑𝑇
and 𝑡=1 𝜆𝑡 = 1. Now, the collective information 𝑏𝑚∗ 𝑛∗ can be
to achieve a higher consensus level, which consists of the following two obtained using a BULIWA operator that factors in the weights of

steps. DMs. Through the comparison of 𝑏𝑚∗ 𝑛∗ and 𝑏𝑡𝑚∗ 𝑛∗ given by 𝐷𝑀𝑡∗ ,
we can derive the following Direction Rules (DRs).

• Identification process DR.1: If 𝑏𝑡𝑚∗ 𝑛∗ > 𝑏𝑚∗ 𝑛∗ , then 𝐷𝑀𝑡∗ should decrease their assess-
This process is utilized to identify the position in 𝑪𝑫 in which ment of the alternative 𝐴𝑚∗ concerning the criterion 𝐶𝑛∗ .

the assessment information given by DMs contributes less to DR.2: If 𝑏𝑡𝑚∗ 𝑛∗ < 𝑏𝑚∗ 𝑛∗ , then 𝐷𝑀𝑡∗ should increase their assess-
achieving a higher consensus level. Identifying the DMs whose ment of the alternative 𝐴𝑚∗ concerning the criterion 𝐶𝑛∗ .

opinion in this position is furthest away from the opinions of DR.3: If 𝑏𝑡𝑚∗ 𝑛∗ = 𝑏𝑚∗ 𝑛∗ , then 𝐷𝑀𝑡∗ does not need to modify their
other DMs is also intended. The position in 𝑪𝑫 that needs to be assessment.

791
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Algorithm 1 The proposed CRP with BULI inputs on the alternative 𝐵𝑀2 with respect to the criterion 𝐴𝐶3 is
( ) { } discovered to make the least contribution and should be adjusted
Input: Decision matrices 𝑩 = 𝑏𝑡𝑚𝑛 𝑀×𝑁 |𝑡 = 1, ⋯ , 𝑇
𝑡
given by DMs
{ } { } to achieve higher consensus level based on Eqs. (6) and (7).
𝐷𝑀1 , ⋯ , 𝐷𝑀𝑇 with the weighting vector 𝜆1 , ⋯ , 𝜆𝑡 , ⋯ , 𝜆𝑇 ; threshold
value 𝛾; maximum number MaxIte. } To this end, the direction process will be activated. Following
{ (of iterations
)
|𝑡 = 1, ⋯ , 𝑇 this process, 𝐷𝑀3 should decrease his or her assessment on
Initialization: 𝐼𝑡𝑒 = 0, 𝑩0𝑡 = 𝑏𝑡𝑚𝑛,0 ⟨ ⟩ ⟨( 𝑃 ) ⟩
𝑀×𝑁 this position since 𝑏323 = 𝑠𝑃5 ; 0.9 > 𝑏23 = 𝑠 , −0.1 ; 0.9 .
1: for 𝑡1 , 𝑡2 = 1, ⋯ , 𝑇 and 𝑡1 ≠ 𝑡2 do ⟨ 𝑃 ⟩ ⟨ 𝑃4 ⟩
2: for 𝑚 = 1, ⋯ , 𝑀 (and 𝑛 = 1, ⋯ ,)𝑁 do 𝐷𝑀3 modifies the judgement from 𝑠5 ; 0.9 to 𝑠4 ; 0.9 , and the
3:
𝑡1 𝑡2
𝑐𝑑𝑚𝑛,𝐼𝑡𝑒
𝑡1
= 𝑠𝑖𝑚 𝑏𝑚𝑛,𝐼𝑡𝑒
𝑡2
, 𝑏𝑚𝑛,𝐼𝑡𝑒 consensus degree on the position is raised to 0.82, meaning that a
4: end for satisfied consensus level has been achieved and the first iteration
5: end for ends.
6: for 𝑚 = 1, ⋯ ,(𝑀 and 𝑛 = 1, ⋯ , 𝑁)/
do Continued from the first iteration, the second iteration will be

𝑇
∑𝑇
𝑡1 𝑡2 activated, and the novel global consensus degree is 0.8, and thus,
7: 𝑐𝑑𝑚𝑛,𝐼𝑡𝑒 = 𝑐𝑑𝑚𝑛,𝐼𝑡𝑒 (𝑇 (𝑇 − 1)∕2)
𝑡1 =1 𝑡2 =1;𝑡2 ≠𝑡1 the global satisfied consensus has been achieved, which ends the
8: end for
{ }
algorithm. After the CRP, the collective assessment information
9: 𝑐𝑑𝐼𝑡𝑒 = min 𝑐𝑑𝑚𝑛,𝐼𝑡𝑒 |𝑚 = 1, ⋯ , 𝑀; 𝑛 = 1, ⋯ , 𝑁 can be elaborated as below:
𝑚,𝑛
10: if 𝑐𝑑𝐼𝑡𝑒 ≥ 𝛾 or 𝐼𝑡𝑒 = MaxIte then ( ) (⟨ ( ) ⟩)
11: Go to Step 26
𝑩 = 𝑏𝑚𝑛 3×3 = 𝛥𝑆 𝜓𝑚𝑛 ; 𝑐𝑚𝑛 3×3
⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩
12: else if 𝑐𝑑𝐼𝑡𝑒 < 𝛾 or 𝐼𝑡𝑒 < MaxIte then
( ) ⎛ 𝑠5 , 0 ; 0.8 𝑠5 , 0.4 ; 0.82 𝑠5 , 1 ; 0.93 ⎞
13: Determine the position 𝐴𝑚∗ ,𝐼𝑡𝑒 , 𝐶𝑛∗ ,𝐼𝑡𝑒 satisfying ⎜⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩⎟
{ } = ⎜ 𝑠5 , −0.4 ; 0.81 𝑠5 , 0.3 ; 0.87 𝑠4 , −0.4 ; 0.9 ⎟
𝑐𝑑𝑚∗ 𝑛∗ ,𝐼𝑡𝑒 = min 𝑐𝑑𝑚𝑛,𝐼𝑡𝑒 |𝑚 = 1, ⋯ , 𝑀; 𝑛 = 1, ⋯ , 𝑁 ⟨( ) ⟩ ⟨( ) ⟩ ⟨( ) ⟩
𝑚,𝑛 ⎜ ⎟
14: Determine the expert 𝐷𝑀𝑡∗ ,𝐼𝑡𝑒 satisfying ⎝ 𝑠𝑃5 , 1 ; 0.72 𝑠𝑃5 , −0.3 ; 0.89 𝑠𝑃5 , 1 ; 0.79 ⎠
{ }
∑𝑇
∗ ∑
𝑇
𝑐𝑑𝑚𝑡 ∗𝑡𝑛∗ ,𝐼𝑡𝑒 = min 𝑐𝑑𝑚𝑙𝑡∗ 𝑛∗ ,𝐼𝑡𝑒 |𝑙 = 1, ⋯ , 𝑇
𝑡=1,𝑡≠𝑡∗ 𝑙 𝑡=1,𝑡≠𝑙
4.3. Determination of ACs and the CIRs using BULI-based QFD
( )
1
15: 𝑏𝑚∗ 𝑛∗ ,𝐼𝑡𝑒 = 𝖡𝖴𝖫𝖨𝖶𝖠 𝑏𝑚∗ 𝑛∗ ,𝐼𝑡𝑒 , ⋯ , 𝑏𝑇𝑚∗ 𝑛∗ ,𝐼𝑡𝑒
16: end if QFD aims to achieve the transformation of CRs into DRs using HOQ
17:

if 𝑏𝑡𝑚∗ 𝑛∗ ,𝐼𝑡𝑒 > 𝑏𝑚∗ 𝑛∗ ,𝐼𝑡𝑒 then so as to be capable of effectively allocating the limited resources to
18: Conduct DR.1 maximize customer satisfaction. In this section, instead of DRs and their

19: else if 𝑏𝑡𝑚∗ 𝑛∗ ,𝐼𝑡𝑒 < 𝑏𝑚∗ 𝑛∗ ,𝐼𝑡𝑒 then CIRs, we introduce the idea of QFD into the determination of ACs and
20: conduct DR.2 their CIRs to guide SBMS. It is worth noting that the determination

21: else if 𝑏𝑡𝑚∗ 𝑛∗ ,𝐼𝑡𝑒 = 𝑏𝑚∗ 𝑛∗ ,𝐼𝑡𝑒 then of CIRs of ACs should consider the relative contribution levels of ACs
22: conduct DR.3 to fulfilling CRs as well as satisfying the basic needs and standards of
23: end if quality and functionality. The detailed steps are illustrated in Fig. 3.
24: 𝐼𝑡𝑒 = 𝐼𝑡𝑒 + 1
25: Go to Step 1.
• Identification of clients and CRs and the specification of IRs
26: for 𝑡 = 1, 𝑚 = 1, ⋯ , 𝑀, 𝑛 = 1, ⋯ , 𝑁 do
of CRs. Aimed at facilitating the incorporation of CRs into SBMS,
27: 𝑏̄ 𝑡𝑚𝑛 = 𝑏𝑡𝑚𝑛,𝐼𝑡𝑒
𝑡 ( ) DMs responsible for SBMS figure out the real clients who can
28: B̄ = 𝑏̄ 𝑡𝑚𝑛 𝑀×𝑁
define the aesthetic and functional requirements for the build-
29: end for { }
𝑡 ( ) ing and further identify the real CRs for building materials to
Output: Decision matrices B̄ = 𝑏̄ 𝑡 𝑚𝑛 𝑀×𝑁 |𝑡 = 1, ⋯ , 𝑇 with a satisfied
be satisfied through surveys, interviews, questionnaires, reviews
consensus level.
or other methods. The clients of a building are individuals or
The steps of the proposed algorithm for CRP can be complied as organizations who fund the project directly or indirectly and gen-
Algorithm 1. The proposed algorithm for CRP is used to ensure erally include the investors, purchasers, occupants, and end-users.
the achievement of satisfied consensus level among DMs before Without loss of generality, let 𝑪𝑳 = {𝐶𝐿1 , … , 𝐶𝐿𝑘 , … , 𝐶𝐿𝐾 }
synthesizing and exploiting collective arguments, and the follow- be the set of clients identified by the decision team. Next, the
ing example continued from Example 1 is used to demonstrate its team prioritizes the clients based on the influence level and the
implementation. related knowledge. The weighting vector obtained can be denoted
{ }
as 𝝀𝐶𝐿 = 𝜆𝐶𝐿 , … , 𝜆𝐶𝐿 , … , 𝜆𝐶𝐿 satisfying 0 ≤ 𝜆𝐶𝐿 ≤ 1 and
Example 2 (Continued from Example 1). The proposed algorithm ∑𝐾 𝐶𝐿 1 𝑘 𝐾 𝑘
𝑘 𝜆𝑘 ={1. Furthermore, we can} denote CRs provided by clients
for CRP should be conducted primarily to preprocess the individ- by 𝑪𝑹 = 𝐶𝑅1 , … , 𝐶𝑅𝑙 , … , 𝐶𝑅𝐿 .
ual assessment information to achieve a satisfied consensus level Identifying the CRs is followed by the determination of the IRs
among the three DMs before synthesizing and distilling collective of CRs which is used to measure the influence levels of CRs on
assessment information. Suppose that the three DMs set consensus enhancing client satisfaction and to guide the determination of
threshold value and maximum number of iterations as 𝛾 = 0.8 and the weights of the ACs. Because of the complexity inherent in
"MaxIte" = 5, respectively. the context and the fuzziness of human cognition and perception,
( ) the clients are asked to give their judgements on the IRs of CRs
During the first iteration, the consensus matrix 𝑪𝑫 = 𝑐𝑑𝑚𝑛 3×3
in the form {of BULI instead of accurate values, which can } be
can be determined in accordance with Eqs. (3) and (4), and the ( )𝑇
result is presented as below: denoted by 𝑪𝑹𝑰 𝑘 = 𝑐𝑟𝑖𝑘1 , … , 𝑐𝑟𝑖𝑘𝑙 , … , 𝑐𝑟𝑖𝑘𝐿 |𝑘 = 1, … , 𝐾 , in
⟨ ( ) ⟩
which 𝑐𝑟𝑖𝑘𝑙 = 𝛥𝑆 𝜓𝑙𝑘 ; 𝑐𝑙𝑘 indicates the IR of 𝐶𝑅𝑙 given by 𝐶𝐿𝑘 ,
⎛ 0.9 0.81 0.95⎞ 𝑘
and 𝑐𝑙 ∈ [0, 1] depicts the reliability degree of this judgement.
𝑪𝑫 = ⎜0.81 0.83 0.64⎟
⎜ ⎟ Before calculating the collective assessment information, we con-
⎝ 0.8 0.86 0.9 ⎠
duct CRP using Algorithm 1 to reach a satisfactory agreement
Then, we can determine the global consensus degree 𝑐𝑑 = 0.64, level among clients. After that, we denote the assessment infor-
{ }
which indicates that the consensus degree is lower than the mation at a satisfied consensus level by 𝑪𝑹𝑰 𝑘 |𝑘 = 1, … , 𝐾 .
threshold value, and the feedback mechanism need to be acti- With the assistance of BULIWA, we can attain the collective IRs
vated in bid to adjust the assessment information. During the vector with weights of clients counted, which is represented by
( )𝑇 ⟨ ( ) ⟩
identification process, the assessment information given by 𝐷𝑀3 𝑪𝑹𝑰 = 𝑐𝑟𝑖1 , … , 𝑐𝑟𝑖𝑙 , … , 𝑐𝑟𝑖𝐿 , in which 𝑐𝑟𝑖𝑙 = 𝛥𝑆 𝜓𝑙 ; 𝑐𝑙 can

792
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Fig. 3. Determination of ACs and their CIRs using BULI-based QFD.

be calculated as follows equation.


⟨ ( ) ⟩
𝑟𝑙𝑛 = 𝛥𝑆 𝜓𝑙𝑛 , 𝑐𝑙𝑛
⟨ ( ) ⟩ ( )
𝑐𝑟𝑖𝑙 = 𝛥𝑆 𝜓𝑙 , 𝑐𝑙 = 𝖡𝖴𝖫𝖨𝖶𝖠 𝑟1𝑙𝑛 , … , 𝑟𝑇𝑙𝑛
( ) (⟨ ( 1 ) 1 ⟩ ⟨ ( 𝑇 ) 𝑇 ⟩)
= 𝖡𝖴𝖫𝖨𝖶𝖠 𝑐𝑟𝑖1𝑙 , … , 𝑐𝑟𝑖𝐾 = 𝖡𝖴𝖫𝖨𝖶𝖠 𝛥𝑆 𝜓𝑙𝑛 ; 𝑐𝑙𝑛 , … , 𝛥𝑆 𝜓𝑙𝑛 ; 𝑐𝑙𝑛
(⟨ ( )
𝑙
⟩ ⟨ ( ) ⟩) ⟨ ( ( 1) ( 𝑇 )) ( 1 𝑇
)⟩ (9)
= 𝖡𝖴𝖫𝖨𝖶𝖠 𝛥𝑆 𝜓𝑙1 ; 𝑐𝑙1 , … , 𝛥𝑆 𝜓𝑙𝐾 ; 𝑐𝑙𝐾 = 𝖳𝖶𝖠 𝛥𝑆 𝜓𝑙𝑛 , … , 𝛥𝑆 𝜓𝑙𝑛 , 𝖶𝖠 𝑐𝑙𝑛 , … , 𝑐𝑙𝑛
⟨ ( ( 1) ( 𝐾 )) ( 1 )⟩ (8) ⟨ 𝑇 ⟩
= 𝖳𝖶𝖠 𝛥𝑆 𝜓𝑙 , … , 𝛥𝑆 𝜓𝑙 , 𝖶𝖠 𝑐𝑙 , … , 𝑐𝑙𝐾 ∑ ∑𝑇
⟨ 𝐾 ⟩ = 𝛥( 𝜆𝐷𝑀𝑡 𝜓𝑙𝑛𝑡
); 𝜆𝐷𝑀
𝑡
𝑡
𝑐𝑙𝑛
∑ ∑
𝐾
𝑡=1 𝑡=1
= 𝛥( 𝜆𝐶𝐿
𝑘 𝜓𝑙
𝑘
); 𝜆 𝐶𝐿 𝑘
𝑘 𝑐 𝑙 ( )
𝑘=1 𝑘=1 where 𝛥𝑆 𝜓𝑙𝑛 indicates the collective assessment information on
the relationship between 𝐴𝐶𝑛 and 𝐶𝑅𝑙 , 𝑐𝑙𝑛 ∈ [0, 1] represents the
collective reliability degree of the knowledge of decision team on
( )
where 𝛥𝑆 𝜓𝑙 represents the collective assessment information on this relationship.
the IR of 𝐶𝑅𝑖 and 𝑐𝑙 ∈ [0, 1] indicates the reliability degree of the • The determination of initial IRs of ACs.
collective judgement. Integrating IRs of CRs with the relationships between CRs and
• The identification of ACs and the determination of the rela- ACs, we can obtain the initial IRs of ACs which can be denoted
( )
tionship between CRs and ACs. After the determination of IRs by 𝑨𝑪𝑰 + = 𝑎𝑐𝑖+ 1
, … , 𝑎𝑐𝑖+
𝑙
, … , 𝑎𝑐𝑖+𝐿
, in which 𝑎𝑐𝑖+ 𝑙
depicts the
of CRs, the decision team is asked to specify a set of ACs, in which initial IR of 𝐴𝐶𝑛 or the relative contribution level made by 𝐴𝐶𝑛 to
some capture the solutions to satisfy CRs, and others serve as the fulfilling the CRs. To achieve this, we should firstly normalize the
{ }
specifications of SBMs. Let 𝑨𝑪 = 𝐴𝐶1 , … , 𝐴𝐶𝑛 , … , 𝐴𝐶𝑁 denote collective IRs of CRs 𝑪𝑹𝑰 and the collective relationship matrix
{ }
the set of ACs, 𝑫𝑴 = 𝐷𝑀1 , … , 𝐷𝑀𝑡 , … , 𝐷𝑀𝑇 denote the de- 𝑹.
{ ⟨ ( ) ⟩
cision team with the weighting vector 𝝀𝐷𝑀 = 𝜆𝐷𝑀 , … , 𝜆𝐷𝑀 ,…, Given that 𝑐𝑟𝑖𝑙 = 𝛥𝑆 𝜓𝑙 , 𝑐𝑙/ ∈ 𝑪𝑹𝑰, we can calculate its
𝐷𝑀
} 𝐷𝑀 ∑ 𝑇 𝐷𝑀
1 𝑡
∑𝐿
𝜆𝑇 satisfying 0 ≤ 𝜆𝑡 ≤ 1 and 𝑡 𝜆𝑡 = 1. While incorporat- normalized form 𝑐𝑟𝑖𝑙 = 𝜓𝑙 𝑐𝑙 𝑙=1 𝜓𝑙 𝑐𝑙 , where 𝑐𝑟𝑖𝑙 ∈ [0, 1]
ing CRs into SBMS, the team members determine the relationships ∑𝐿
and
{ 𝑙=1 𝑐𝑟𝑖 𝑙 = } 1. Then the normalized IRs of CRs 𝑪𝑹𝑰 =
between CRs and ACs by analyzing to what extent each AC can
𝑐𝑟𝑖𝑙 |𝑙 = 1, … , 𝐿 are obtained. Similarly, given 𝑟𝑙𝑛 =
relate to and influence each CR. To facilitate the elicitation of the ⟨ ( ) ⟩
relationship information with inherent uncertain and qualitative 𝛥𝑆 𝜓𝑙𝑛 , 𝑐𝑙𝑛 ∈ 𝑹, its normalized form can be determined 𝑟̄𝑙𝑛 =
∑ ∑𝑁
characteristics, the members can utilize the LTS 𝑆 𝑅 to depict the 𝜓𝑙𝑛 𝑐𝑙𝑛 ∕ 𝑁𝑛=1 𝜓𝑙𝑛 𝑐𝑙𝑛 , in which 𝑟̄𝑙𝑛 ∈ [0, 1] and 𝑛=1 𝑟̄𝑙𝑛(= 1. Now we
)
relationship information in the form of BULI, and the relationship can derive the normalized relationship matrix 𝑹 = 𝑟̄𝑙𝑛 𝐿×𝑁 with
information can form a matrix with the CRs as rows and the the sum of each row being one.
After the normalization, we can calculate the initial IRs of ACs
ACs{as columns. Thus, we can}denote the relationship matrices
( ) ⟨ ( 𝑡) 𝑡 ⟩ ( )
by 𝑹𝑡 = 𝑟𝑡𝑙𝑛 𝐿×𝑁 |𝑡 = 1, … , 𝑇 , in which 𝑟𝑡𝑙𝑛 = 𝛥𝑆 𝜓𝑙𝑛 ; 𝑐𝑙𝑛 𝑨𝑪𝑰 + = 𝑎𝑐𝑖+ 1
, … , 𝑎𝑐𝑖+
𝑙
, … , 𝑎𝑐𝑖+
𝐿
, in which 𝑎𝑐𝑖+ 𝑙
can be calculated
depicts the influence level of 𝐴𝐶𝑛 toward 𝐶𝑅𝑙 given by 𝐷𝑀𝑡 , and by the following equation.
𝑡 ∈ [0, 1] indicates the reliability degree of this judgement.
𝑐𝑙𝑛 ∑
𝐿
Analogously, before synthesizing the individual relationship ma- 𝑎𝑐𝑖+
𝑛 = 𝑐𝑟𝑖𝑙 𝑟̄𝑙𝑛 (10)
𝑙=1
trices, CRP using Algorithm 1 is performed to reach a high
∑𝑁
agreement level among DMs on the relationship between CRs and where 𝑎𝑐𝑖+ 𝑛 ∈ [0, 1] and
+
𝑛=1 𝑎𝑐𝑖𝑛 = 1.
ACs. Normally, we can also { denote the relationship matrices
} with • Determination of modified IRs of ACs.
( )
satisfied consensus level 𝑹𝑡 = 𝑟𝑡𝑙𝑛 𝐿×𝑁 |𝑡 = 1, … , 𝑇 . Next, the The CIR of an AC describes its relative importance or influence
( )
collective relationship matrix 𝑹 = 𝑟𝑙𝑛 𝐿×𝑁 is obtained taking level on SBMS. It not only reflects its relative contribution level to
weighting vector 𝝀𝐷𝑀 into consideration with the utilization of fulfilling CRs, but also covers its relative
BULIWA, in which 𝑟𝑙𝑛 = ⟨𝑠𝑙𝑛 ; 𝑐𝑙𝑛 ⟩ can be attained by the following importance in conforming to the basic standards or needs of

793
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

quality and function of this type of building material. Therefore, an aggregating process using BULIWA, we can derive the collective
before achieving the CIRs of ACs, the decision team should assessment information with a satisfied consensus level, which can be
( ) ⟨ ( ) ⟩
provide the modified IRs of ACs utilizing the LTS 𝑆 𝐼 in the denoted by 𝑩 = 𝑏𝑚𝑛 𝑀×𝑁 , in which 𝑏𝑚𝑛 = 𝛥𝑆 𝜓𝑚𝑛 , 𝑐𝑚𝑛 can be
form of BULI. The modified importance ratings are denoted as calculated by the following formula
{ ( ) }
𝑨𝑪𝑰 ∗𝑡 = 𝑎𝑐𝑖∗𝑡 , … , 𝑎𝑐𝑖∗𝑡 , … , 𝑎𝑐𝑖∗𝑡 |𝑡 = 1, … , 𝑇 , in which 𝑎𝑐𝑖∗𝑡𝑛 = ⟨ ( ) ⟩
⟨ ( ∗𝑡 ) ∗𝑡 ⟩1 ( ∗𝑡𝑛) 𝑁
𝑏𝑚𝑛 = 𝛥𝑆 𝜓𝑚𝑛 ; 𝑐𝑚𝑛
𝛥𝑆 𝜓𝑛 ; 𝑐𝑛 , 𝛥𝑆 𝜓𝑛 indicates the modified importance rat- ( )
ing of criterion 𝐴𝐶𝑛 given by 𝐷𝑀𝑡 and 𝑐𝑛∗𝑡 represents the reliabil- = 𝖡𝖴𝖫𝖨𝖶𝖠 𝑏1𝑚𝑛 , … , 𝑏𝑇𝑚𝑛
(⟨ ( 1 ) 1 ⟩ ⟨ ( 𝑇 ) 𝑇 ⟩)
ity degree of the DM’s perception of the criterion. = 𝖡𝖴𝖫𝖨𝖶𝖠 𝛥𝑆 𝜓𝑚𝑛 ; 𝑐𝑚𝑛 , … , 𝛥𝑆 𝜓𝑚𝑛 ; 𝑐𝑚𝑛
Algorithm 1 for CRP within BULI context is performed first to ⟨ ( ( 1 ) ( 𝑇 )) ( 1 𝑇
)⟩ (14)
= 𝖳𝖶𝖠 𝛥𝑆 𝜓𝑚𝑛 , … , 𝛥𝑆 𝜓𝑚𝑛 , 𝖶𝖠 𝑐𝑚𝑛 , … , 𝑐𝑚𝑛
reach a preferable version with satisfied consensus level which is ⟨ 𝑇 ⟩
{ ( ) }
also denoted by 𝑨𝑪𝑰 ∗𝑡 = 𝑎𝑐𝑖∗𝑡 ∑ ∑
𝑇
𝑛 , … , 𝑎𝑐𝑖𝑁 |𝑡 = 1, … , 𝑇 .
, … , 𝑎𝑐𝑖∗𝑡 ∗𝑡
1 = 𝛥( 𝜆𝐷𝐸 𝑡
𝑡 𝑠𝑚𝑛 ); 𝜆𝐷𝐸 𝑡
𝑡 𝑐𝑚𝑛
Next, we aggregate the individual opinions on the modified IRs 𝑡=1 𝑡=1
into a collective one with the assistance of BULIWA and weighting ( )
(
vector 𝝀𝐷𝑀 , which is denoted by 𝑨𝑪𝑰 ∗ = 𝑎𝑐𝑖∗1 , … , 𝑎𝑐𝑖∗𝑛 , … , where 𝛥𝑆 𝜓𝑚𝑛 represents the collective judgement on the performance
)
𝑎𝑐𝑖∗𝑁 , where 𝑎𝑐𝑖∗𝑛 can be obtained by the following formula. of alternative 𝐵𝑀𝑚 with respect to criterion 𝐴𝐶𝑛 , 𝑐𝑚𝑛 indicates the
⟨ ( ) ⟩ collective reliability degree of the assessment information given by the
𝑎𝑐𝑖∗𝑛 = 𝛥𝑆 𝜓𝑛∗ ; 𝑐𝑛∗ decision team.
( ∗1 ) ( )
= 𝖡𝖴𝖫𝖨𝖶𝖠 𝑎𝑐𝑖𝑛 , … , 𝑎𝑐𝑖∗𝑇 𝑛 Once the collective decision matrix 𝑩 = 𝑏𝑚𝑛 𝑀×𝑁 has been ob-
(⟨ ( ) ⟩ ⟨ ( ) ⟩)
= 𝖡𝖴𝖫𝖨𝖶𝖠 𝛥𝑆 𝜓𝑛∗1 ; 𝑐𝑛∗1 , … , 𝛥𝑆 𝜓𝑛∗𝑇 ; 𝑐𝑛∗𝑇 tained, it should be exploited to determine the most suitable SBM.
⟨ ( ( ∗1 ) ( ∗𝑇 )) ( ∗1 )⟩ (11) ELECTRE III is therefore performed, and it depends highly on the
= 𝖳𝖶𝖠 𝛥𝑆 𝜓𝑛 , … , 𝛥𝑆 𝜓𝑛 , 𝖶𝖠 𝑐𝑛 , … , 𝑐𝑛∗𝑇
⟨𝑇 ⟩ construction of binary outranking relations of each pair of alternatives.
∑ ∑
𝑇
For any pair of alternative SBM 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 ∈ 𝑩𝑴, we can denote
∗𝑡 𝐷𝑀 ∗𝑡
= 𝜆𝐷𝑀
𝑡 𝜓𝑛 ; 𝜆 𝑡 𝑐𝑛
𝑡=1 𝑡=1
the binary outranking relations by 𝐵𝑀𝑚1 𝐵𝑀𝑚2 which indicates 𝐵𝑀𝑚1
( ) is at least as good as 𝐵𝑀𝑚2 . To demonstrate whether the assertion
In the above formula, 𝛥𝑆 𝜓𝑛∗ describes the collective modified 𝐵𝑀𝑚1 𝐵𝑀𝑚2 is true, we need to determine the concordance and
IRs of criterion 𝐴𝐶𝑛 and 𝐶𝑛∗ represents the reliability degree of discordance indices on each criterion to weigh the strength of support
the information possessed by the decision team on this criterion. and opposition of each criterion, which requires the assistance of indif-
Furthermore, the modified IRs of ACs are( normalized with the
) ference threshold 𝑞𝑛 , preference threshold 𝑝𝑛 and veto threshold 𝑣𝑛 on
∗ ∗ ∗ ∗
normalized version denoted as 𝐀𝐂𝐈 = 𝑎𝑐𝑖1 , … , 𝑎𝑐𝑖𝑛 , … , 𝑎𝑐𝑖𝑁 , each criterion. If there are sufficient criteria to affirm this assertion and

where 𝑎𝑐𝑖𝑛 can be calculated by the following equation with there is no criterion to strongly refute this assertion, then the assertion
∗ ∑ ∗
𝑎𝑐𝑖𝑛 ∈ [0, 1] and 𝑁
𝑛=1 𝑎𝑐𝑖𝑛 = 1.
is true. Otherwise, the assertion is not true. Combining the concordance
and discordance indices, we can derive the credibility degree of the
∗ 𝜓 ∗𝑐∗
𝑎𝑐𝑖𝑛 = ∑𝑁 𝑛 𝑛 . (12) assertion 𝐵𝑀𝑚1 𝐵𝑀𝑚2 . The detailed exploitation steps are as follows.
∗ ∗
𝑛=1 𝜓𝑛 𝑐𝑛

• Determination of CIRs of ACs. • Concordance index. ( )


Combining the initial IRs of ACs and the modified IRs of ACs, we Concordance index 𝜌 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 is introduced to depict the
(
can obtain the CIRs which is denoted by 𝑨𝑪𝑰 = 𝑎𝑐𝑖1 , … , 𝑎𝑐𝑖𝑛 , strength of the evidence provided by all the criteria for sup-
)
… , 𝑎𝑐𝑖𝑁 , in which 𝑎𝑐𝑖𝑛 indicates the comprehensive IR of 𝐴𝐶𝑛 porting the assertion 𝐵𝑀𝑚1 𝐵𝑀𝑚2 . To achieve it, we firstly
( )
and can be calculated by the following formula determine the partial concordance index 𝜌𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 on
( )

𝑎𝑐𝑖𝑛 = 𝜇 + 𝑎𝑐𝑖+ ∗
(13) each criterion. As for the benefit criteria, 𝜌𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 can
𝑛 + 𝜇 𝑎𝑐𝑖𝑛
∑𝑁 be calculated as
where 𝑎𝑐𝑖𝑛 ∈ [0, 1], 𝑛=1 𝑎𝑐𝑖𝑛 = 1, 𝜇+ , 𝜇∗ ∈ [0, 1], 𝜇 + + 𝜇∗ = 1,
( )
and 𝜇 + represents the relative influence levels of the initial IRs 𝜌𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2
on the building materials selection, and 𝜇 ∗ indicates those of the ⎧1 ( )
⎪ if 𝑏𝑚1 𝑛 ≥ 𝑏𝑚2 𝑛 or 𝑏𝑚1 𝑛 < 𝑏𝑚2 𝑛 and 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ≤ 𝑞𝑛
modified IRs on the building materials selection. ( )
⎪ 𝑝 − 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ( )
=⎨ 𝑛 if 𝑞𝑛 ≤ 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ≤ 𝑝𝑛 and 𝑏𝑚1 𝑛 < 𝑏𝑚2 𝑛
⎪ 𝑝𝑛 − 𝑞𝑛
( )
4.4. Assessment and exploitation processes using BULI-based ELECTRE III ⎪0 if 𝑝𝑛 ≤ 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 and 𝑏𝑚1 𝑛 < 𝑏𝑚2 𝑛

(15)
This subsection elaborates the steps involved in the assessment
( )
and exploitation processes. Before aggregating individual opinions, the As for cost criteria, 𝜌𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 can be calculated by the
algorithm for CRP needs to be conducted to ensure the integrity of
following formula
the aggregated collective assessment information with a satisfactory
( )
consensus level among DMs. 𝜌𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2
We can denote the set of alternative building materials by 𝑩𝑴 = ⎧1 ( )
{ } { if 𝑏𝑚1 𝑛 ≤ 𝑏𝑚2 𝑛 or 𝑏𝑚1 𝑛 > 𝑏𝑚2 𝑛 and 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ≤ 𝑞𝑛
𝐵𝑀1 , … , 𝐵𝑀𝑚 , … , 𝐵𝑀𝑀 . Given the decision team 𝑫𝑴 = 𝐷𝑀1 , ⎪ ( )
} { } ⎪ 𝑝 − 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ( )
… , 𝐷𝑀𝑡 , … , 𝐷𝑀𝑇 and the set of ACs 𝑨𝑪 = 𝐴𝐶1 , … , 𝐴𝐶𝑛 , … , 𝐴𝐶𝑁 , =⎨ 𝑛 if 𝑞𝑛 ≤ 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ≤ 𝑝𝑛 and 𝑏𝑚1 𝑛 > 𝑏𝑚2 𝑛
⎪ 𝑝𝑛 − 𝑞𝑛
each DM is asked to articulate their judgements on the performances of ( )
⎪0 if 𝑝𝑛 ≤ 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 and 𝑏𝑚1 𝑛 > 𝑏𝑚2 𝑛

alternatives on criteria in the form of BULI with the assistance of the
{ ( ) (16)
LTS 𝑆 𝑃 , which can form a decision matrix denoted as 𝑩 𝑡 = 𝑏𝑡𝑚𝑛 𝑀×𝑁
⟨ ( 𝑡 ) 𝑡 ⟩
|𝑡 = 1, … , 𝑇 }, where 𝑏𝑚𝑛 = 𝛥𝑆 𝜓𝑚𝑛 , 𝑐𝑚𝑛 indicates the performance
𝑡 ( )
value of alternative 𝐵𝑀𝑚 on criterion 𝐴𝐶𝑛 given by 𝐷𝑀𝑡 and 𝑐𝑚𝑛 𝑡 ∈ In the above two equations, 𝜌𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 ∈ [0, 1] depicts
[0, 1] depicts the reliability degree of this judgement. the strength of the evidence provided by criterion 𝐴𝐶𝑛 for sup-
porting the assertion 𝐵𝑀𝑚1 𝐵𝑀𝑚2 . Next, the overall concordance
After the elicitation of assessment information, the algorithm for ( )
CRP depicted in Fig. 2 is performed to examine or modify the consensus index 𝜌 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 can be calculated by synthesizing the
level among DMs until a satisfied consensus level is achieved. Through partial concordance indices using traditional weighted arithmetic

794
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

averaging operator, which can be written as … , 𝑀), can be calculated as

( ) ∑𝑁 ( ) ( ) ∑𝑀 ( )
𝜌 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 = 𝑎𝑐𝑖𝑛 𝜌𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 (17) 𝜑− 𝐵𝑀𝑚 = 𝜎 𝐵𝑀𝑚1 , 𝐵𝑀𝑚 (22)
𝑛=1 𝑚1 =1

in which 𝑎𝑐𝑖𝑛 is the CIR of 𝐴𝐶𝑛 . • Net credibility degree of each SBM outranking other SBMs.
• Discordance index for each criterion. Combining the concordant credibility degree with the discor-
In addition, the strength of evidence against this assertion on dant credibility degree, the net credibility degree of each SBM
each criterion is introduced to ensure that none of the criteria outranking other SBMs can be determined as
is strongly against this assertion. The main reason for this is that ( ) ( ) ( )
𝜑 𝐵𝑀𝑚 = 𝜑+ 𝐵𝑀𝑚 − 𝜑− 𝐵𝑀𝑚 , (23)
very bad performance in one criterion may not ( be compensated
) ( )
by good performance in another. we let 𝛽𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 de- where 𝜑 𝐵𝑀𝑚 represents the net strength of the domination of
note the partial discordance index, which depicts the strength other alternatives by the alternative 𝐵𝑀𝑚 . Then, all the BMs can
of(evidence provided
) by criterion 𝐴𝐶𝑛 for refusing
( the assertion) be ranked in accordance with the net credibility degree, as the
𝜌 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 . As for the benefit criteria, 𝛽𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 higher the net credibility degree, the better the BM.
can be calculated as
( )
5. Case study
𝛽𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2
⎧1 ( ) In this section, the proposed model is applied to a case study on
⎪ ( if 𝑏𝑚1 𝑛 < 𝑏𝑚2 𝑛 and 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ≥ 𝑣𝑛 ,
)
⎪ 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 − 𝑝𝑛 ( ) the assessment and selection of office floorings. Firstly, the problem
=⎨ if 𝑏𝑚1 𝑛 < 𝑏𝑚2 𝑛 and 𝑝𝑛 ≤ 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ≤ 𝑣𝑛 ,
⎪ 𝑣𝑛 − 𝑝𝑛 formulation of office flooring selection is presented. Next, the proposed
( )
⎪0 if 𝑏𝑚1 𝑛 ≥ 𝑏𝑚2 𝑛 or 𝑏𝑚1 𝑛 < 𝑏𝑚2 𝑛 and 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ≤ 𝑝𝑛 . model is employed to assess alternatives and determine the ranking

of alternatives. Finally, both sensitivity and comparative analyses are
(18)
conducted to demonstrate the feasibility and rationality of the proposed
( ) model.
As for cost criteria, 𝛽𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 can be calculated by the
following formula
5.1. Structuring of office flooring selection problem
( )
𝛽𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2
⎧1 ( ) The selection of office flooring occupies a significant role in office
⎪ ( if 𝑏𝑚1 𝑛 > 𝑏𝑚2 𝑛 and 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ≥ 𝑣𝑛 ,
) design, as the floorings not only provide their basic functionalities but
⎪ 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 − 𝑝𝑛 ( )
=⎨ if 𝑏𝑚1 𝑛 > 𝑏𝑚2 𝑛 and 𝑝𝑛 ≤ 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ≤ 𝑣𝑛 , also contribute to the aesthetics which can reflect company image and
⎪ 𝑣𝑛 − 𝑝𝑛
( ) values and inspire the productivity and creativity of the employees.
⎪0 if 𝑏𝑚1 𝑛 ≤ 𝑏𝑚2 𝑛 or 𝑏𝑚1 𝑛 > 𝑏𝑚2 𝑛 and 𝑑 𝑏𝑚1 𝑛 , 𝑏𝑚2 𝑛 ≤ 𝑝𝑛 .
⎩ By this we mean the office flooring selection not only conforms with
(19) the values and aesthetic style of the company, but also considers
the many factors impacting the functionalities and comfort, including
• Determining the credibility degrees of outranking relations. quiet, high-level traffic, durability, healthy, green, ease of cleaning
Once the determination of concordance and discordance indices and maintenance, less cost, and so on. To this end, there are multiple
have been completed, we need to calculate the credibility degree objectives that need to be achieved and balanced amid the selection
of the assertion 𝐵𝑀𝑚1 𝐵𝑀𝑚2 by combining the concordance and process, which inevitably increases the complexity of office flooring
discordance indices using( the following
) formula (20) which is selection. Besides, many factors are affecting the selection of office
given in Box I, where 𝐼𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 is a set of criteria such floorings, including the specific use of the space to be used, subfloor
( ) ( )
that 𝛽𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 > 𝜌 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 . composition and structure, the functional requirements for office floor-
ing, and the desired aesthetics. Since these factors are rooted in CRs
The credibility degrees
( ( for all pairs))of alternatives can form a that are central to office design, CRs need to be incorporated into office
credibility matrix 𝜎 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 . To achieve the final flooring selection, which adds to the complexity.
𝑀×𝑀
ranking of the alternatives, we introduce an easier method than Furthermore, in the flooring market, there are many flooring types
the distillation methods to further exploit the credibility matrix. available to select from, e.g., rubber flooring, luxury vinyl planks,
Referring to the literature (Hashemi et al., 2016) the net credi- carpet tiles, solid hardwood flooring, laminate flooring, which poses
bility degree of each SBM outranking other SBMs, which reflects even further difficulties in the selection process. While dealing with the
the relative comprehensive performance of each alternative, can complexities mentioned above, the decision team is nudged to handle
be calculated by the following steps. the office flooring selection problem in the use of a scientific systematic
• Concordant credibility degree of each SBM outranking other MCGDM method, which incorporates CRs into the selection process and
SBMs. copes effectively with multiple alternatives and conflicting criteria. This
( ( ))
Depending on the credibility matrix, 𝜎 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 , scenario has inspired the application of the comprehensive decision
𝑀×𝑀 model developed in this proposal to the selection process of commercial
the concordant credibility degree of each SBM outranking other office flooring.
( )
SBMs, 𝜑+ 𝐵𝑀𝑚 (𝑚 = 1, … , 𝑀), can be calculated by the follow- The case study involves a technology company that is devoted
ing equation to the creation of a sustainable and comfortable workspace for its
employees. The workspace is also intended to be used for the reception
( ) ∑𝑀 ( )
𝜑+ 𝐵𝑀𝑚 = 𝜎 𝐵𝑀𝑚 , 𝐵𝑀𝑚1 (21) of its customers and stakeholders occasionally. Therefore, the company
𝑚1 =1 desires that the workspace to be developed should not only provide
comfortable, green, and healthy working conditions to inspire better
• Discordant credibility degree of each SBM outranked by concentration and productivity of employees, but also to promote an
other SBMs. image of sustainability to visitors. In order to achieve these goals,
Depending on the credibility matrix, the discordant credibility the company is required to seek assistance from an interior design
( )
degree of each SBM outranked by other SBMs, 𝜑− 𝐵𝑀𝑚 (𝑚 = 1, firm for finding solutions. In the event, the firm builds a decision

795
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

( ) ( ) ( )
⎧ 𝜌 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 if 𝛽𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 ≤ 𝜌 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 ,
⎪ ( )
( ) ⎪
𝜎 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 = ⎨ ( ) ∏ 1 − 𝛽𝑛 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 (20)
⎪𝜌 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 ( ) otherwise.
( )
⎪ 𝑛∈𝐼𝑛 𝐵𝑀𝑚 ,𝐵𝑀𝑚 1 − 𝜌 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2
⎩ 1 2

Box I.

Table 1
The assessment information on IRs of CRs elicited by clients.
𝐶𝑅1 𝐶𝑅2 𝐶𝑅3 𝐶𝑅4 𝐶𝑅5 𝐶𝑅6 Consensus degree
⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩
𝐶𝐿1 𝑠 ; 0.8 𝑠 ; 0.7 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9
⟨ 𝐼2 ⟩ ⟨ 𝐼3 ⟩ ⟨ 𝐼4 ⟩ ⟨ 𝐼5 ⟩ ⟨ 𝐼6 ⟩ ⟨ 𝐼1 ⟩
𝐶𝐿2 𝑠 ;1 𝑠 ; 0.7 𝑠4 ; 0.8 𝑠5 ; 0.8 𝑠6 ; 1 𝑠 ;1
⟨ 𝐼2 ⟩ ⟨ 𝐼3 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼1 ⟩ 0.88 > 0.8
𝐶𝐿3 𝑠2 , 1 𝑠4 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠3 ; 1
⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼3 ⟩ ⟨ 𝐼4 ⟩ ⟨ 𝐼6 ⟩ ⟨ 𝐼 ⟩
𝐶𝐿4 𝑠 ;1 𝑠 ;1 𝑠4 ; 0.6 𝑠4 ; 1 𝑠6 ; 1 𝑠 ; 0.8
⟨ 𝐼3 ⟩ ⟨ 𝐼4 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼3 ⟩
𝐶𝐿5 𝑠3 ; 0.8 𝑠3 ; 0.7 𝑠4 ; 0.6 𝑠4 ; 1 𝑠6 ; 1 𝑠1 ; 0.7
⟨( 𝐼 ) ⟩ ⟨( 𝐼 ) ⟩ ⟨( 𝐼 ) ⟩ ⟨( 𝐼 ) ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩
Collective preference 𝑠2 , 0.4 ; 0.94 𝑠3 , 0.5 ; 0.8 𝑠4 , −0.25 ; 0.75 𝑠4 , 0.35 ; 0.96 𝑠6 ; 1 𝑠2 ; 0.89

team consisting of four designers responsible for the office design. resistance (𝐴𝐶4 ), slip resistance (𝐴𝐶5 ), thickness swelling or water
{ }
The DMs are denoted by the set 𝐷𝑀1 , 𝐷𝑀2 , 𝐷𝑀3 , 𝐷𝑀4 with the absorption (𝐴𝐶6 ), and stain resistance (𝐴𝐶7 ); (ii) sustainable criteria
weighting vector (0.2, 0.35, 0.2, 0.25) determined as per their knowledge, that consist of aesthetic appeal (𝐴𝐶8 ), formaldehyde emission (𝐴𝐶9 ),
experience, and design skills. The designers first need to appreciate the volatile organic compounds emission (𝐴𝐶10 ), thermal resistance (𝐴𝐶11 ),
various CRs to familiarize themselves with the problem and understand and ability to be recycled (𝐴𝐶12 ); and (iii) economic criteria that mainly
their preferences with the aesthetic style and various functionalities refer to market price (𝐴𝐶13 ). These criteria are used as basis for the
to be achieved through interviews and surveys. After discussion, the assessment of alternative office flooring designs and the selection of
designers and clients come to a joint decision as to the overall aesthetic the optimal one. The specific implications of each AC are offered in
style and functionalities of the office, which would act as a guide during Table 2. Once the ACs are specified, the decision team is required to
the office design process. The office is occupied by company employees utilize BULI pairs to evaluate the relationships between CRs and ACs.
who spend a great deal of their time and energy every day on various The resulting relationship matrices have been presented in Table 3.
creative tasks. Thus, they want the office to provide them with an open, Following the elicitation of assessment information, the algorithm for
healthy, and warm space for them to focus on their work and to relieve CRP under BULI is carried out to achieve satisfactory consensus among
stress. From the viewpoint of the employers, they want the designers to the team members. During the first iteration of the CRP, the global con-
bear in mind the sustainable image, aesthetic style, duration, and cost. sensus degree calculated is 0.747, being lower than the threshold value.
As such, the feedback mechanism is activated to adjust the assess-
5.2. Determination of ACs and their CIRs for office flooring selection ment information. During the adjustment procedure, an identification
process is performed primarily to identify the position at which the
Fulfilling the requirements of employees and employers for office assessment information needs to be adjusted and the DM responsible for
floorings selection necessitates the decision team to specify the CRs adjustment. The result reveals that the designer 𝐷𝑀1 needs to change
accurately and incorporates CRs into the selection process to determine their assessment information regarding the relationship between 𝐶𝑅6
ACs and their CIRs by using the two-phase QFD methodology. The team and 𝐴𝐶2 . After the identification, the direction process is performed to
selects five clients as sources of requirements, including employees, guide the modification. The result suggests the designer 𝐷𝑀2 should
⟨ ⟩ ⟨ ⟩
employers, and visitors of the company. The clients are denoted as
{ } modify the original assessment from the value 𝑠𝑅 , 1 to 𝑠𝑅 , 0.7 on
⟨ 𝑅 ⟩ 0 1
𝐶𝐿1 , 𝐶𝐿2 , 𝐶𝐿3 , 𝐶𝐿4 , 𝐶𝐿5 and the corresponding weighting vector is the position because 𝑠0 , 1 given by 𝐷𝑀1 on the position is less than
⟨( ) ⟩
(0.15, 0.2, 0.25, 0.25, 0.15). Subsequently, the team specifies the CRs for the collective assessment value 𝑠𝑅 2
, −0.3 , 0.93 , which is determined
the office flooring. The CRs include ease of cleaning and maintenance by aggregating the individual assessment information on the position
(𝐶𝑅1 ), durability (𝐶𝑅2 ), quietness (𝐶𝑅3 ), style and comfort (𝐶𝑅4 ), using a BULIWA operator. The completion of the adjustment promotes
sustainability (𝐶𝑅5 ), and cost-effectiveness (𝐶𝑅6 ). The clients are then the consensus degree on the position to 0.818, indicating a satisfied
requested to prioritize them in order of preference. Given the incom- consensus level has been achieved and ends the first iteration. The sec-
pleteness of information and the inherent uncertainty of the cognition, ond iteration is then activated. The global consensus degree obtained
the assessment information on the IRs of CRs are elicited in the form becomes 0.75, a figure lower than the threshold value, indicating the
of BULI paris on the basis of the LTS 𝑆 𝐼 , which is presented in Table 1. necessity of adjustment. In a similar fashion, after the identification and
Before the determination of the collective IRs of CRs, Algorithm 1 for direction processes, 𝐷𝑀1 is requested to reassess their assessment on
CRP is carried out to achieve a satisfied agreement among clients. The the relationship between 𝐶𝑅2 and 𝐴𝐶13 . This changes the assessment
⟨ ⟩ ⟨ ⟩
decision team sets consensus threshold value and the maximum number from 𝑠𝑅 0
, 1 to 𝑠𝑅1
, 1 . At that moment, the consensus degree on the
of iterations as 𝛾 = 0.8 and 𝙼𝚊𝚡𝙸𝚝𝚎 = 5, respectively. During the position gets raised to 0.83, implying the achievement of a satisfactory
first iteration of CRP, the consensus degree is 0.88, being greater than consensus level on the position. Following the second iteration, the
the threshold value. In this sense, a satisfied consensus level has been third iteration is activated and the global consensus reaches the degree
achieved such that the follow-up steps can continue. The IRs of CRs of 0.818. The global satisfactory consensus has been achieved, which
obtained after the congregation of the individual preferences are listed ends the implementation of the algorithm.
in Table 1.
To capture the CRs for office flooring, the decision team specifies After completing CRP, the aggregation of individual relationship
three aspects of criteria: (i) functional criteria that include acous- matrices is conducted with the aid of the BULIWA operator and the
tic properties (𝐴𝐶1 ), hardness (𝐴𝐶2 ), wear resistance (𝐴𝐶3 ), impact weights of DMs, and the collective relationship matrix with satisfied

796
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Table 2
The implications of the ACs for SBMS.
Types Criteria Implications References
Acoustic properties (𝐴𝐶1 ) This criterion was introduced to weigh noise level of flooring, and Jeon et al. (2006)
it depends highly on the material of flooring. In an office setting, a
type of flooring with low performance in this criterion would
impact the mood and productivity of employees and employers
negatively. This suggested that the flooring to be selected should
perform better on this criterion.
Hardness (𝐴𝐶2 ) In general, hardness was used to measure the density of hardwood Hosking (2017), Meyer et al.
flooring and is one of the important indicators to measure the (2011) and Bustos et al.
durability of hardwood flooring. However, hardness was introduced (2009)
to weigh softness and comfortability of the flooring in this paper,
mainly because the wear and impact resistance rather than
hardness, was used to weigh durability and the alternative floorings
include not only hardwood flooring, but also rubber flooring, carpet
flooring, ceramic tiles, etc., which will be outlined below. Thus, it
is a cost criterion.
Wear resistance (𝐴𝐶3 ) Wear resistance refers to the capability of the flooring to stand up James (1985)
to harsh conditions and heavy traffic. It is a benefit criterion. A
flooring design with higher performance in this criterion can keep
flooring attractive for many years without having to repair or
replace.
Functional criteria Impact resistance (𝐴𝐶4 ) Impact resistance is the ability of the flooring to withstand a high Harrison and Brough (1992)
force or shock applied to it over a short period of time. The
flooring with high performance in this criterion can avoid holes,
gouges, and indentations in tough environments. Besides, wear
resistance and impact resistance are two important aspects of
durability. It is a benefit criterion.
Slip resistance (𝐴𝐶5 ) The surface of a flooring with slip resistance has a higher Harris and Shaw (1988) and
coefficient of friction, which can help prevent the employees and James (1985)
visitors slipping on the flooring. It is also a benefit criterion.
Thickness swelling or water absorption This criterion refers to how much the flooring might swell in a Gu et al. (2005)
(𝐴𝐶6 ) wet environment. The better the performance in this criterion, the
worse water resistance of the flooring is, and the shorter service life
of the flooring in wet environment. This is a cost criterion.
Stain resistance (𝐴𝐶7 ) Stain resistance is the ability to withstand discoloration caused by Harrison and Brough (1992),
contact with liquids or materials. High performance of a flooring Stonhard (2018) and Colletti
according to this criterion points to its ease of maintenance and et al. (1998)
reduction of lifetime cost. This is a benefit criterion.
Aesthetic appeal (𝐴𝐶8 ) Aesthetic appeal has different meaning for different objects. In an Eagleton (1990) and Adorno
office flooring selection setting, a flooring with a wide range of (1997)
design options or natural beauty performs better on this criterion.
Hence, it is also a benefit criterion.
Formaldehyde emission (𝐴𝐶9 ) Formaldehyde is a colorless, flammable, strong-smelling chemical Wiglusz et al. (2002) and
gas. Exposure to formaldehyde for a long time may pose critical Perratore (2016)
impacts on health. Typically, the gas usually appears in building
materials, such as floorings, paint, and furniture, etc. In SBMS, this
criterion is introduced to weigh the emission level of formaldehyde
and is a cost criterion.
Sustainable criteria Volatile organic compounds VOCs are organic chemicals and exist in the form of gas at room Wiglusz et al. (2002) and
(VOCs) emission (𝐴𝐶10 ) temperature. They include a variety of chemicals except Brown et al. (1994)
formaldehyde, most of which have adverse health effects. Therefore,
VOCs emission is also a cost criterion.
Thermal resistance (𝐴𝐶11 ) Thermal resistance is the ability of a flooring to resist the flow of Seo et al. (2011)
heat through the flooring and depends on the thermal properties of
flooring materials. It plays an important role in determining energy
efficiency. Hence, it is a benefit criterion.
Ability to be recycled (𝐴𝐶12 ) This criterion refers to the ability of flooring to be recycled and Level (2018) and Nicoletti
depends on the material of flooring. It contributes to sustainability, et al. (2002)
so it is a benefit criterion.
Economic criteria Market price (𝐴𝐶13 ) Market price refers to the price of office flooring and is the main Mousavi-Nasab and
but not the only aspect of cost-effectiveness. Cost-effectiveness Sotoudeh-Anvari (2018b)
mainly refers to the life-cycle cost including market price and
maintenance cost. Furthermore, market price is a cost criterion.

consensus level is given in Table 4. Subsequently afterwards, the IRs Subsequently, the modified IRs of ACs need to be determined.
of CRs and the collective relationship matrix are normalized, and the Based on the LTS 𝐒𝐼 , each DM presents the assessment information
results are presented in Table 5. Then, integrating the normalized IRs on the modified IRs of ACs in the form of BULI to depict the relative
of CRs with the normalized relationship matrix between CRs and ACs importance of each AC in satisfying the basic standards of quality and
( )
gives the initial IRs of ACs 𝑨𝑪𝑰 + = 𝑎𝑐𝑖+ 1
, 𝑎𝑐𝑖+
2
, … , 𝑎𝑐𝑖+
13
, which has functionality. The assessment information is presented in Table 6. Then
also been included in Table 5. the algorithm for CRP is carried out to achieve a satisfied agreement
among DMs on the modified IRs. During the first iteration of the

797
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Table 3
Relationship matrices between CRs and ACs given by DMs.
𝐴𝐶1 𝐴𝐶2 𝐴𝐶3 𝐴𝐶4 𝐴𝐶5 𝐴𝐶6 𝐴𝐶7 𝐴𝐶8 𝐴𝐶9 𝐴𝐶10 𝐴𝐶11 𝐴𝐶12 𝐴𝐶13
⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩
𝐶𝑅1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 5𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩
𝐶𝑅2 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 5𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩
𝐷𝑀1 𝐶𝑅3 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 5𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩
𝐶𝑅4 𝑠 ;1 𝑠 ; 0.6 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 2𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 2𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩
𝐶𝑅5 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠3 ; 1 𝑠 ;1 𝑠6 ; 1 𝑠 ;1 𝑠 ;1 𝑠0 ; 1
⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 𝑅 ⟩
𝐶𝑅6 𝑠0 ; 1 𝑠0 ; 1 𝑠4 ; 0.8 𝑠0 ; 1 𝑠2 ; 0.8 𝑠0 ; 1 𝑠3 ; 0.8 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠2 ; 1 𝑠2 ; 0.8 𝑠6 ; 1
⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩
𝐶𝑅1 𝑠0 ; 1 𝑠0 ; 1 𝑠 ;1 𝑠4 ; 0.7 𝑠 ;1 𝑠4 ; 1 𝑠 ;1 𝑠 ;1 𝑠0 ; 1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩
𝐶𝑅2 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 5𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩
𝐷𝑀2 𝐶𝑅3 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 5𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 2𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩
𝐶𝑅4 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩
𝐶𝑅5 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠0 ; 1 𝑠2 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 2𝑅 ⟩
𝐶𝑅6 𝑠0 ; 1 𝑠3 ; 0.8 𝑠5 ; 0.6 𝑠0 ; 1 𝑠1 ; 1 𝑠0 ; 1 𝑠4 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠3 ; 0.8 𝑠3 ; 1 𝑠6 ; 1
⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩
𝐶𝑅1 𝑠0 ; 1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠0 ; 1 𝑠6 ; 1 𝑠6 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠 ;1 𝑠0 ; 1 𝑠0 ; 1
⟨ 𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩
𝐶𝑅2 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8
⟨ 0𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 5𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 5𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 3𝑅 ⟩
𝐷𝑀3 𝐶𝑅3 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8
⟨ 5𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 2𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩
𝐶𝑅4 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.7 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩
𝐶𝑅5 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 2𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 2𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 2𝑅 ⟩
𝐶𝑅6 𝑠0 ; 1 𝑠2 ; 1 𝑠4 ; 0.6 𝑠0 ; 1 𝑠2 ; 1 𝑠0 ; 1 𝑠4 ; 0.6 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠2 ; 1 𝑠1 ; 1 𝑠6 ; 1
⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩
𝐶𝑅1 𝑠0 ; 1 𝑠0 ; 1 𝑠 ;1 𝑠4 ; 1 𝑠0 ; 1 𝑠5 ; 1 𝑠 ;1 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1
⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩
𝐶𝑅2 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 2𝑅 ⟩
𝐷𝑀4 𝐶𝑅3 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 5𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩
𝐶𝑅4 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.6 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩
𝐶𝑅5 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8
⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 1𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 3𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 4𝑅 ⟩ ⟨ 6𝑅 ⟩ ⟨ 2𝑅 ⟩
𝐶𝑅6 𝑠0 ; 1 𝑠1 ; 1 𝑠3 ; 0.6 𝑠0 ; 1 𝑠1 ; 1 𝑠0 ; 1 𝑠5 ; 0.7 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠1 ; 1 𝑠2 ; 0.8 𝑠6 ; 1

Table 4
Relationship matrices between CRs and ACs given by DMs.
𝐴𝐶1 𝐴𝐶2 𝐴𝐶3 𝐴𝐶4 𝐴𝐶5 𝐴𝐶6 𝐴𝐶7
⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨( 𝑅 ) ⟩ ⟨( 𝑅 ) ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩
𝐶𝑅1 𝑠0 ; 1 𝑠0 ; 1 𝑠 , −0.35 ; 0.94 𝑠4 , −0.4 ; 0.86 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅4 ⟩ ⟨( 𝑅 ) ⟩ ⟨ 0𝑅 ⟩ ⟨(4.85
𝑅
) ⟩ ⟨(6 𝑅 ) ⟩
𝐶𝑅2 𝑠0 ; 1 𝑠4 ; 0.96 𝑠6 ; 1 𝑠 , 0.4 ; 1 𝑠0 ; 1 𝑠 , −0.05 ; 0.93 𝑠 , 0.35 ; 0.95
⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅4 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅4 ⟩ ⟨ 𝑅3 ⟩
𝐶𝑅3 𝑠 ;1 𝑠 ;1 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠 ;1 𝑠 ;1
⟨ 5𝑅 ⟩ ⟨(0 𝑅 ) ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨( 𝑅 ) ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩
𝐶𝑅4 𝑠 ;1 𝑠 , −0.05 ; 0.85 𝑠0 ; 1 𝑠 ;1 𝑠 , −0.35 ; 1 𝑠0 ; 1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 𝑅1 ⟩ ⟨ 𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨( 1𝑅 ) ⟩ ⟨ 𝑅 ⟩ ⟨ 0𝑅 ⟩
𝐶𝑅5 𝑠 ;1 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠2 , −0.45 ; 0.88 𝑠0 ; 1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨( 𝑅 ) ⟩ ⟨( 𝑅 ) ⟩ ⟨ 𝑅 ⟩ ⟨( 𝑅 ) ⟩ ⟨ 𝑅 ⟩ ⟨(0 𝑅 ) ⟩
𝐶𝑅6 𝑠0 ; 1 𝑠2 , −0.1 ; 0.87 𝑠4 , 0.1 ; 0.64 𝑠0 ; 1 𝑠1 , 0.4 ; 0.96 𝑠0 ; 1 𝑠4 , 0.05 ; 0.81
𝐴𝐶8 𝐴𝐶9 𝐴𝐶10 𝐴𝐶11 𝐴𝐶12 𝐴𝐶13
⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩
𝐶𝑅1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨(0 𝑅 ) ⟩
𝐶𝑅2 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠0 ; 1 𝑠 ;1 𝑠 , −0.35 ; 0.96
⟨(0 𝑅 ) ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨( 2𝑅 ) ⟩
𝐶𝑅3 𝑠 , 0.45 ; 1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠1 , −0.45 ; 0.96
⟨ 𝑅2 ⟩ ⟨ 0𝑅 ⟩ ⟨ 0𝑅 ⟩ ⟨(0 𝑅 ) ⟩ ⟨ 0𝑅 ⟩ ⟨( 𝑅 ) ⟩
𝐶𝑅4 𝑠 ;1 𝑠0 ; 1 𝑠0 ; 1 𝑠 , 0.05 ; 0.77 𝑠0 ; 1 𝑠 , 0.35 ; 1
⟨(6 𝑅 ) ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅3 ⟩ ⟨ 𝑅 ⟩ ⟨( 0𝑅 ) ⟩
𝐶𝑅5 𝑠 , −0.2 ; 0.93 𝑠6 ; 1 𝑠6 ; 1 𝑠 ;1 𝑠6 ; 1 𝑠 , −0.4 ; 0.95
⟨ 𝑅3 ⟩ ⟨ 𝑅 ⟩ ⟨ 𝑅 ⟩ ⟨(4 𝑅 ) ⟩ ⟨( 𝑅 ) ⟩ ⟨ 𝑅2 ⟩
𝐶𝑅6 𝑠0 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠2 , 0.1 ; 0.93 𝑠2 , 0.15 ; 0.91 𝑠6 ; 1

Table 5
The normalized IRs of CRs, the normalized relationship matrix, the initial IRs of ACs, the normalized modified IRs of ACs and the CIRs of ACs.
IRs of CRs 𝐴𝐶1 𝐴𝐶2 𝐴𝐶3 𝐴𝐶4 𝐴𝐶5 𝐴𝐶6 𝐴𝐶7 𝐴𝐶8 𝐴𝐶9 𝐴𝐶10 𝐴𝐶11 𝐴𝐶12 𝐴𝐶13
𝐶𝑅1 0.114 0.000 0.000 0.198 0.177 0.000 0.279 0.346 0.000 0.000 0.000 0.000 0.000 0.000
𝐶𝑅2 0.141 0.000 0.164 0.257 0.188 0.000 0.196 0.135 0.000 0.000 0.000 0.000 0.000 0.060
𝐶𝑅3 0.142 0.627 0.000 0.000 0.000 0.000 0.000 0.000 0.307 0.000 0.000 0.000 0.000 0.066
𝐶𝑅4 0.211 0.000 0.080 0.000 0.000 0.064 0.000 0.000 0.591 0.000 0.000 0.231 0.000 0.034
𝐶𝑅5 0.303 0.000 0.000 0.000 0.000 0.049 0.000 0.000 0.095 0.218 0.218 0.146 0.218 0.055
𝐶𝑅6 0.090 0.000 0.084 0.140 0.000 0.072 0.000 0.174 0.000 0.000 0.000 0.104 0.105 0.321
𝑨𝑪𝑰 + 0.800 0.089 0.048 0.071 0.047 0.035 0.059 0.074 0.197 0.066 0.066 0.102 0.075 0.071
Relative influence level
𝑨𝑪𝑰 ∗ 0.200 0.034 0.121 0.112 0.095 0.065 0.113 0.085 0.046 0.096 0.095 0.052 0.058 0.028
𝑨𝑪𝑰 0.078 0.062 0.079 0.056 0.041 0.070 0.076 0.167 0.072 0.072 0.092 0.072 0.062

algorithm, the consensus obtained is equal to 0.93, implying that a modified IRs of ACs are normalized, the results of which is expressed
satisfied consensus level has been achieved and the algorithm ends. in Table 5.
And then the aggregation process is conduct to determine the collective
( )
modified IRs of ACs and the results 𝑨𝑪𝑰 ∗ = 𝑎𝑐𝑖∗1 , 𝑎𝑐𝑖∗2 , … , 𝑎𝑐𝑖∗13 The decision team then sets the values of relative influence levels
are also presented in Table 6. Following the aggregation process, the of initial IRs and modified IRs as 0.8 and 0.2, respectively. Then, we
combine the initial IRs of ACs with the modified IRs of ACs to determine

798
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Table 6
Modified IRs of ACs provided by individuals and the collective outputs.
𝐴𝐶1 𝐴𝐶2 𝐴𝐶3 𝐴𝐶4 𝐴𝐶5 𝐴𝐶6 𝐴𝐶7
⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩
𝐷𝐸1 𝑠 ;1 𝑠 ; 0.9 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.7 𝑠 ;1
⟨ 𝐼1 ⟩ ⟨ 𝐼5 ⟩ ⟨ 𝐼5 ⟩ ⟨ 𝐼4 ⟩ ⟨ 𝐼3 ⟩ ⟨ 𝐼5 ⟩ ⟨ 𝐼4 ⟩
𝐷𝐸2 𝑠1 ; 1 𝑠 ;1 𝑠 ; 0.8 𝑠4 ; 0.8 𝑠 ;1 𝑠 ; 0.7 𝑠4 ; 0.8
⟨ 𝐼 ⟩ ⟨ 𝐼5 ⟩ ⟨ 𝐼5 ⟩ ⟨ 𝐼 ⟩ ⟨ 3𝐼 ⟩ ⟨ 𝐼5 ⟩ ⟨ 𝐼 ⟩
𝐷𝐸3 𝑠 ; 0.7 𝑠5 ; 0.8 𝑠6 ; 0.7 𝑠 ; 0.7 𝑠 ; 0.8 𝑠5 ; 1 𝑠 ;1
⟨ 𝐼2 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼5 ⟩ ⟨ 𝐼3 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼3 ⟩
𝐷𝐸4 𝑠 ; 0.9 𝑠 ; 0.8 𝑠 , 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠3 ; 1
⟨(2 ) ⟩ ⟨(6 𝐼 ) ⟩ ⟨(6 𝐼 ) ⟩ ⟨(4 𝐼 ) ⟩ ⟨(2 𝐼 ) ⟩ ⟨(6 𝐼 ) ⟩ ⟨( 𝐼 ) ⟩
Collective outputs 𝑠𝐼1 , 0.45 ; 0.92 𝑠5 , 0.25 ; 0.89 𝑠5 , 0.2 , 0.83 𝑠4 , 0.2 ; 0.87 𝑠3 , −0.25 ; 0.92 𝑠5 , 0.25 ; 0.84 𝑠4 , −0.45 ; 0.93
𝐴𝐶8 𝐴𝐶9 𝐴𝐶10 𝐴𝐶11 𝐴𝐶12 𝐴𝐶13
⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 𝐼 ⟩
𝐷𝐸1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8
⟨ 2𝐼 ⟩ ⟨ 4𝐼 ⟩ ⟨ 4𝐼 ⟩ ⟨ 𝐼2 ⟩ ⟨ 𝐼2 ⟩ ⟨ 𝐼1 ⟩
𝐷𝐸2 𝑠 ;1 𝑠 ; 0.7 𝑠 ; 0.8 𝑠2 ; 1 𝑠 ;1 𝑠 ;1
⟨ 2𝐼 ⟩ ⟨ 5𝐼 ⟩ ⟨ 4𝐼 ⟩ ⟨ 𝐼 ⟩ ⟨ 2𝐼 ⟩ ⟨ 1𝐼 ⟩
𝐷𝐸3 𝑠2 ; 1 𝑠 ; 0.8 𝑠 ; 0.7 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9
⟨ 𝐼 ⟩ ⟨ 𝐼4 ⟩ ⟨ 5𝐼 ⟩ ⟨ 𝐼2 ⟩ ⟨ 3𝐼 ⟩ ⟨ 1𝐼 ⟩
𝐷𝐸4 𝑠 ; 0.7 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.7
⟨ 2 ⟩ ⟨(4 𝐼 ) ⟩ ⟨(4 𝐼 ) ⟩ ⟨ 𝐼2 ⟩ ⟨(3 𝐼 ) ⟩ ⟨(2 𝐼 ) ⟩
Collective outputs 𝑠𝐼2 ; 0.89 𝑠4 , 0.35 ; 0.86 𝑠4 , 0.2 ; 0.87 𝑠2 ; 1 𝑠2 , 0.45 ; 0.91 𝑠1 , 0.25 ; 0.87

( )
the CIRs of ACs. The results 𝑨𝑪𝑰 = 𝑎𝑐𝑖1 , 𝑎𝑐𝑖2 , … , 𝑎𝑐𝑖13 have been at a new ranking of office flooring building based on the proposed
included in Table 5. model. In the sensitivity analysis of CRs, there are 24 experiments to be
performed (see Fig. 4(a)). The ranking of office floorings is sensitive to
5.3. Assessment and exploitation of alternative office floorings changes in the IRs of CRs. There are also more conspicuous fluctuations
of rankings on the CRs with higher IRs, e.g., 𝐶𝑅1 , 𝐶𝑅3 , 𝐶𝑅4 , and
Upon the determination of ACs and their corresponding CIRs, the 𝐶𝑅5 . Meanwhile, the changing IRs of CRs poses an impact on the
DMs choose 8 types of alternative office floorings available from the final rankings of alternatives as the CIRs of ACs have been taken into
target market, which includes rubber flooring tiles (𝐵𝑀1 ), luxury vinyl account. This process expands the range of influence posed by the
planks (𝐵𝑀2 ), carpet tiles (𝐵𝑀3 ), solid hardwood flooring (𝐵𝑀4 ), en- changing IRs of CRs through the relationship matrix.
gineered wood flooring (𝐵𝑀5 ), laminate flooring (𝐵𝑀6 ), ceramic tiles Examining the sensitivity of the final ranking of alternatives on the
or porcelain tiles (𝐵𝑀7 ), and terrazzo (𝐵𝑀8 ). For more information on changing CIRs of ACs is also required for our study. Therefore, a sim-
these alternatives, please refer to Table 7. Based on the LTS 𝑆 𝑃 , the ilar experimental method is performed on the ACs. Meanwhile, other
DMs assess the performance of each alternative on each AC in the form parameters are kept the same. In the event, there are 52 experiments
{ ( ) }
of BULI. The assessment information, 𝐁𝑡 = 𝑏𝑡𝑚𝑛 8×13 |𝑡 = 1, … , 4 , is to conduct (see Fig. 4(b)). The results show that the rankings of office
presented in Table 8. The assessment constructions are followed with an floorings are also sensitive to changes in terms of the CIRs of ACs. We
execution of the algorithm for CRP to achieve a satisfactory agreement discover at the same time that the sensitivity level of the final ranking
among the DMs. During the CRP, the fact that the global consensus on an AC is related to its CIR of the AC and the deviation among the
degree is lower than the threshold value leads to a series of adjust- performances of alternatives. As for the former, a criterion with higher
ments. For instance, the assessment information 𝑏327 given by 𝐷𝑀3 is CIR has a significant influence on the final ranking of alternatives,
⟨ ⟩ ⟨ ⟩
adjusted from 𝑠𝑃6 , 1 to 𝑠𝑃5 , 1 while the assessment information 𝑏2310 which indicates that the sensitivity levels of the criteria are higher, such
⟨ ⟩ ⟨ ⟩
given by 𝐷𝑀2 is adjusted from 𝑠𝑃4 , 1 to 𝑠𝑃3 , 1 . These adjustments as with 𝐴𝐶8 , 𝐴𝐶11 , and 𝐴𝐶3 . Otherwise, the sensitivity levels of criteria
contribute to a satisfactory consensus level. The resulting collective with lower comprehensive IR are lower, e.g., as with 𝐴𝐶5 , 𝐴𝐶4 , and
( )
decision matrix 𝑩 = 𝑏𝑚𝑛 8×13 is presented in Table 9. ELECTRE III 𝐴𝐶2 . As for the latter, if the deviation between the performances of any
can be further introduced to exploit the collective assessment infor- two BMs on a criterion is high, the outranking relation between the pair
mation to obtain the final ranking of alternatives. The concordance is stable even if the CIR of the criterion is high. For example, though the
and discordance indices of each pair of floorings on each criterion deviation between the performances of any two BMs on criterion 𝐴𝐶11
are(determined in))advance by the DMs. Then, the credibility matrix
( with CIR 0.092 is higher, the sensitivity of rankings on this criterion
𝜎 𝐵𝑀𝑚1 , 𝐵𝑀𝑚2 can be determined accordingly (see Table 10). is lower than that on the criteria 𝐴𝐶12 with CIR 0.072 and 𝐴𝐶7 with
8×8 CIR 0.76. Furthermore, the criteria 𝐴𝐶9 , 𝐴𝐶10 and 𝐴𝐶12 have equal
Finally, the net credibility degrees of each office flooring are evaluated CIR values, i.e., 0.072. On the other hand, the sensitivity level on the
in accordance with the credibility matrix (see Table 10). Depending on criterion 𝐴𝐶12 is higher than that of the criteria 𝐴𝐶9 and 𝐴𝐶10 by reason
the net credibility degree of each office flooring, their final ranking can of the higher deviation level of the performances among alternatives on
be determined and is given as follows: the criterion 𝐴𝐶12 .
𝐵𝑀1 ≻ 𝐵𝑀3 ≻ 𝐵𝑀4 ≻ 𝐵𝑀8 ≻ 𝐵𝑀5 ≻ 𝐵𝑀7 ≻ 𝐵𝑀2 ≻ 𝐵𝑀6 . In summary, the sensitivity analysis shows that the IRs of CRs or
the CIRs of ACs have a significant impact on the ranking of office
5.4. Sensitivity analysis floorings because CRs and ACs point out the direction and lay the
foundation for the SBMS. Thus, during SBMS, DMs can be expected
to accurately specify the CRs and the corresponding IRs, and further
The CRs and ACs are subjected to sensitivity analyses, conducted
precisely transfer CRs and their IRs to ACs and the corresponding CIRs
respectively to explore the influence of the IRs of CRs and the CIRs
with the aid of a scientific tool, which necessitates the introduction of
of ACs on the final ranking of office floorings. To achieve this, we
relationship matrix in QFD.
increase the IR of each CR by 30% and 60% respectively and then
decrease them by 30% and 60%, respectively. When the IR of each
CR is increased or decreased, the IRs of other CRs are also changed
5.5. Comparative analysis
accordingly in proportion such that the sums of IRs of CRs equal to 1.
As for the adjustments of each CR, we acquire four new groups of IRs
of CRs. For each group, the comprehensive IRs of ACs are adjusted in In this subsection, we will report the comparative analysis for BULI,
accordance with the HOQ-based analysis process depicted in Section CRP under BULI, QFD-based analysis tool, and ELECTRE III-based
3.3. Besides, other parameters including the weights of DMs and the ranking method to verify the feasibility and rationality of the proposed
thresholds on each criterion are retained. For each group, we arrive MCGDM model.

799
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Table 7
The alternative flooring types available for selection.
Flooring types Picture Content

Rubber flooring tiles Rubber flooring is a sustainable flooring which is made from a
rubber tree. It is available in a variety of colors, patterns, and
textures. It is a type of resilient flooring that comes with inherent
advantages, such as durability, easy maintenance, softness,
water-resistance, and being sound-proof. However, they also have a
series of drawbacks such as high-cost, staining, and anaturally dull
finish.

Source: http://www.remprubberflooring.com/en/references/362-
lund-slaatto-architects-as

Luxury vinyl planks (LVPs) LVP is a type of flooring made from polyvinyl chloride and can
imitate the texture and colors of natural materials such as wood
and stone. It is available in a wide variety of colors and patterns
and possess the following characteristics: water, scuff, stain, and
scratch resistance. Thus, it has extreme durability and is easy to
maintain. However, it may release dangerous emissions indoors,
thus leading to a marginally bad effect on resale value.

Source:
https://www.armstrongflooring.com/commercial/en-cl/project-
photo-galleries/armstrong-custom-build-hospitality.html#photo22

Carpet tiles This is a textile floor covering typically consisting of an upper layer
of pile attached to a backing. It comes in different styles and
patterns and contributes to comfort, warmth, health, quiet and cost.
However, it can easily get stained and require a high level of
maintenance. It also has a shorter lifespan in comparison to other
flooring choices.

Source: https://www.forbo.com/flooring/en-
us/products/flotex/flotex-linear/b62jf4

Solid hardwood flooring This is constructed with strips or planks that are completely
hardwood lumber and has the following advantages: natural look
and warm feel, durability and long-lasting, easy to clean and
maintain, sustainable and healthy. However, it possesses the
following drawbacks: noise, occasional refinishing, high price,
sensitivity to moisture.

Source: https://officesnapshots.com/offices/?browse_page=2&filter_
industry=food-beverage

(continued on next page)

5.5.1. Comparative analysis between the inclusion and exclusion of BULI to form a two-dimensional information representation model. As for
traditional fuzzy linguistic approaches, DMs are limited to elicit assess-
With the combination of BUI and fuzzy linguistic approach, BULI is ment information using only a certain linguistic term. By contrast, BULI
established to express uncertain assessment information and in many provides DMs with a more rich and elaborative approach under a high
ways precedes BUI, as it introduces linguistic variables instead of degree of uncertainty. To verify the effectiveness of the introduction of
precise numbers to facilitate the elicitation of uncertain assessment the degree of reliability, we removed the dimension of the degree of
information approximating human cognitive processes. Compared with { ( ) }
reliability from the decision matrices 𝑩 𝑡 = 𝑏𝑡𝑚𝑛 8×13 |𝑡 = 1, … , 4 , so
traditional fuzzy linguistic approaches which use linguistic variables the assessment information in the matrices degenerates to traditional
to model linguistic information, BULI introduces a novel dimension linguistic variables. We input the adjusted assessment information into
called the degree of reliability to reflect the reliability level of the the proposed model and achieve a novel ranking of office floorings
elicitation of assessment information or the reliability level of the expressible as 𝐵𝑀1 ≻ 𝐵𝑀4 ≻ 𝐵𝑀8 ≻ 𝐵𝑀7 ≻ 𝐵𝑀5 ≻ 𝐵𝑀2 ≻ 𝐵𝑀3 ≻
related knowledge of the DMs. It also accompanies linguistic terms

800
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Table 7 (continued).
Flooring types Picture Content

Engineered wood flooring This is made of several layers of wood/plywood glued and
laminated together and is an exciting alternative to solid hardwood
flooring, implying that it shares most advantages of solid hardwood
flooring. It can withstand both heat and moisture and has a lower
cost than solid hardwood flooring. However, it cannot be refinished
or can only be refinished two or three times. Further, it is
considerably more expensive than laminate and carpet.

Source:
http://nbinteriorstudio.com/catalogue/product/michael-angelo/

Laminate flooring This is a type of synthetic flooring that is made by fusing several
layers of different material together. It is easier to install, durable,
stain resistant and cost effective. However, it is susceptible to
moisture damage and can’t be refinished. It also comes with an
unnatural look and hence has a lower resale value.

Source: https://www.pinterest.com/pin/443182419563570625/

Ceramic or porcelain tiles Both are made of dry clay that is compressed and fired at extreme
temperatures for longer periods to make it dense and strong. They
are durable, water-resistant, stain-resistant and wear-resistant, and
have low maintenance. However, they have some drawbacks. For
example, installing them is difficult, time-consuming and of
high-cost. They are also very slippery when wet.

Source: http://padronflooring.com/gallery/

Terrazzo This is a composite construction material made by combining chips


of aggregates such as marble or stone with concrete or epoxy-resin.
The main advantages are durability, aesthetic appeal, green and
low maintenance. The main drawbacks are high-cost, cold, hard
and the need for professional installation.

Source: https://www.tmsupply.com/Gallery.aspx?subcategory=
Kalama%20River%20Office%20Building

𝐵𝑀6 . Evidently, there is a clear gap between the novel ranking and the the CIRs of ACs remain the same. Then, to obtain the final ranking
{ ( ) }
original one, which highlights the significance of taking the reliability of office floorings, the decision matrices 𝑩 𝑡 = 𝑏𝑡𝑚𝑛 8×13 |𝑡 = 1, … , 4
dimension into account. given by DMs are synthesized directly by the BULIWA operator and
then exploited in the use of ELECTRE III to arrive at the final ranking
5.5.2. Comparative analysis between the inclusion and exclusion of CRP in of office floorings. The following relationship is discovered: 𝐵𝑀1 ≻
the proposed MCGDM model 𝐵𝑀4 ≻ 𝐵𝑀8 ≻ 𝐵𝑀3 ≻ 𝐵𝑀5 ≻ 𝐵𝑀7 ≻ 𝐵𝑀2 ≻ 𝐵𝑀6 . Comparing these
CRP has been introduced in this paper to obtain the final result with rankings with the original one, we could see that the order of 𝐵𝑀3
an acceptable consensus level. If we exclude CRP during the MCGDM, falls from the second to the fourth mainly because of the influence of
abnormal assessment information given by DMs may make a direct abnormal assessment value given by 𝐷𝑀2 on the performance of 𝐵𝑀3
impact on the final outcomes and further lead to neglecting opinions or with respect to 𝐴𝐶10 , which is against the assessments of 𝐷𝑀1 , 𝐷𝑀3
concerns of the majority. To further verify the necessity for introducing and 𝐷𝑀4 . The comparison results highlight the usefulness of CRP in
CRP, we remove CRP from the proposed MCGDM model. We find that the achievement of efficient outputs.

801
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Table 8
Assessment information given by DMs about the performances of alternatives on criteria.
𝐴𝐶1 𝐴𝐶2 𝐴𝐶3 𝐴𝐶4 𝐴𝐶5 𝐴𝐶6 𝐴𝐶7 𝐴𝐶8 𝐴𝐶9 𝐴𝐶10 𝐴𝐶11 𝐴𝐶12 𝐴𝐶13
⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩
𝐵𝑀1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9 𝑠 ; 0.8 𝑠 ; 0.9 𝑠 ; 0.8
⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩
𝐵𝑀2 𝑠 ; 0.7 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.9 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1
⟨ 4𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩
𝐵𝑀3 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8
⟨ 6𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩
𝐷𝑀1 𝐵𝑀4 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.7 𝑠 ; 0.7 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8
⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩
𝐵𝑀5 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.6 𝑠 ; 0.6 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8
⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩
𝐵𝑀6 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.7 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8
⟨ 2𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩
𝐵𝑀7 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.6 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1
⟨ 4𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩
𝐵𝑀8 𝑠5 ; 0.8 𝑠6 ; 0.8 𝑠5 ; 1 𝑠4 ; 0.6 𝑠3 ; 0.8 𝑠1 ; 1 𝑠6 ; 1 𝑠6 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠1 ; 1 𝑠3 ; 0.9 𝑠6 ; 0.9
⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩
𝐵𝑀1 𝑠 ;1 𝑠 ;1 𝑠3 ; 1 𝑠 ; 0.8 𝑠 ; 0.8 𝑠4 ; 1 𝑠4 ; 1 𝑠4 ; 0.8 𝑠2 ; 0.8 𝑠2 ; 1 𝑠4 ; 0.8 𝑠 ; 0.7 𝑠 ;1
⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩
𝐵𝑀2 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1
⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩
𝐵𝑀3 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1
⟨ 6𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 2𝑃 ⟩
𝐷𝑀2 𝐵𝑀4 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.7 𝑠 ; 0.9 𝑠 ; 0.9 𝑠 ;1 𝑠 ; 0.7 𝑠 ;1 𝑠 ;1 𝑠0 ; 1 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1
⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩
𝐵𝑀5 𝑠 ; 0.6 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.9 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1
⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩
𝐵𝑀6 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.7 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9
⟨ 2𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 3𝑃 ⟩
𝐵𝑀7 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.9 𝑠1 ; 1 𝑠1 ; 0.9 𝑠 ;1 𝑠 ;1
⟨ 4𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩
𝐵𝑀8 𝑠4 ; 1 𝑠5 ; 1 𝑠5 ; 1 𝑠2 ; 1 𝑠1 ; 1 𝑠1 ; 1 𝑠5 ; 0.8 𝑠6 ; 1 𝑠1 ; 0.9 𝑠1 ; 1 𝑠1 ; 0.9 𝑠3 ; 1 𝑠5 ; 1
⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩
𝐵𝑀1 𝑠5 ; 0.8 𝑠4 ; 0.8 𝑠3 ; 1 𝑠4 ; 1 𝑠5 ; 0.8 𝑠4 ; 0.8 𝑠 ;1 𝑠4 ; 1 𝑠 ;1 𝑠2 ; 0.9 𝑠 ; 0.8 𝑠5 ; 1 𝑠5 ; 1
⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩
𝐵𝑀2 𝑠 ; 0.6 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.9 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.9
⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩
𝐵𝑀3 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ; 0.8 𝑠2 ; 1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8
⟨ 6𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩
𝐷𝑀3 𝐵𝑀4 𝑠 ;1 𝑠 ;1 𝑠 ; 0.7 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩
𝐵𝑀5 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.5 𝑠 ; 0.9 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1
⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩
𝐵𝑀6 𝑠 ;1 𝑠 ; 0.8 𝑠3 ; 1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.7 𝑠 ;1 𝑠 ;1 𝑠3 ; 1 𝑠3 ; 1 𝑠 ;1 𝑠3 ; 1 𝑠 ;1
⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 3𝑃 ⟩
𝐵𝑀7 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 4𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩
𝐵𝑀8 𝑠4 ; 1 𝑠6 ; 0.8 𝑠6 ; 1 𝑠3 ; 1 𝑠3 ; 1 𝑠0 ; 1 𝑠6 ; 1 𝑠6 ; 1 𝑠0 ; 1 𝑠0 ; 1 𝑠1 ; 1 𝑠3 ; 1 𝑠6 ; 1
⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩
𝐵𝑀1 𝑠5 ; 0.7 𝑠 ;1 𝑠3 ; 0.8 𝑠4 ; 1 𝑠4 ; 1 𝑠3 ; 0.8 𝑠4 ; 1 𝑠5 ; 1 𝑠2 ; 0.9 𝑠2 ; 0.9 𝑠4 ; 0.8 𝑠5 ; 0.8 𝑠5 ; 1
⟨ 𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩
𝐵𝑀2 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1
⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩
𝐵𝑀3 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 6𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩
𝐷𝑀4 𝐵𝑀4 𝑠 ;1 𝑠 ;1 𝑠 ; 0.7 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 4𝑃 ⟩
𝐵𝑀5 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1
⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 0𝑃 ⟩ ⟨ 6𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩
𝐵𝑀6 𝑠 ;1 𝑠 ; 0.8 𝑠3 ; 0.8 𝑠 ; 0.8 𝑠3 ; 1 𝑠 ;1 𝑠4 ; 1 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠2 ; 1 𝑠 ;1
⟨ 2𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 2𝑃 ⟩
𝐵𝑀7 𝑠 ;1 𝑠 ; 0.8 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠 ;1 𝑠 ; 0.9 𝑠 ;1 𝑠 ;1 𝑠 ;1
⟨ 4𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 2𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 5𝑃 ⟩ ⟨ 4𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 1𝑃 ⟩ ⟨ 3𝑃 ⟩ ⟨ 3𝑃 ⟩
𝐵𝑀8 𝑠4 ; 1 𝑠5 ; 1 𝑠5 ; 1 𝑠3 ; 1 𝑠2 ; 1 𝑠1 ; 0.9 𝑠5 ; 1 𝑠5 ; 1 𝑠1 ; 1 𝑠1 ; 1 𝑠1 ; 1 𝑠3 ; 1 𝑠5 ; 1

Table 9
Collective assessment information about the performances of alternatives on criteria.
𝐴𝐶1 𝐴𝐶2 𝐴𝐶3 𝐴𝐶4 𝐴𝐶5 𝐴𝐶6 𝐴𝐶7
⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩ ⟨ 𝑃 ⟩ ⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩ ⟨ 𝑃 ⟩
𝐵𝑀1 𝑠 , −0.5 ; 0.8 𝑠 , −0.4 ; 1 𝑠 ;1 𝑠 , −0.2 ; 0.9 𝑠 , −0.2 ; 0.9 𝑠 , −0.2 ; 0.9 𝑠 ;1
⟨( 𝑃5 ) ⟩ ⟨( 𝑃4 ) ⟩ ⟨(3 𝑃 ) ⟩ ⟨( 𝑃4 ) ⟩ ⟨( 𝑃5 ) ⟩ ⟨( 𝑃4 ) ⟩ ⟨(4 𝑃 ) ⟩
𝐵𝑀2 𝑠 , 0.2 ; 0.9 𝑠 , −0.4 ; 1 𝑠 , 0.4 ; 0.9 𝑠 , −0.5 ; 1 𝑠 , 0.1 ; 0.9 𝑠 , −0.2 ; 1 𝑠 , 0.2 ; 1
⟨ 𝑃4 ⟩ ⟨ 𝑃4 ⟩ ⟨( 𝑃3 ) ⟩ ⟨( 5𝑃 ) ⟩ ⟨ 𝑃3 ⟩ ⟨ 𝑃2 ⟩ ⟨( 4𝑃 ) ⟩
𝐵𝑀3 𝑠6 ; 1 𝑠2 ; 1 𝑠 , −0.5 ; 0.8 𝑠 , −0.2 ; 1 𝑠6 ; 1 𝑠5 ; 0.9 𝑠 , 0.4 ; 0.9
⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨( 3𝑃 ) ⟩ ⟨ 𝑃3 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃2 ⟩
𝐵𝑀4 𝑠 ; 0.9 𝑠 ; 0.9 𝑠 , −0.2 ; 0.7 𝑠 ; 0.9 𝑠 ; 0.9 𝑠 ; 0.9 𝑠 ; 0.8
⟨ 𝑃1 ⟩ ⟨ 𝑃3 ⟩ ⟨( 𝑃6 ) ⟩ ⟨ 𝑃5 ⟩ ⟨ 𝑃5 ⟩ ⟨(3 𝑃 ) ⟩ ⟨ 5𝑃 ⟩
𝐵𝑀5 𝑠 ; 0.8 𝑠 ; 0.9 𝑠 , 0.4 ; 0.8 𝑠5 ; 0.9 𝑠 ; 0.9 𝑠 , −0.4 ; 1 𝑠 ; 0.9
⟨ 𝑃1 ⟩ ⟨(3 𝑃 ) ⟩ ⟨ 𝑃4 ⟩ ⟨( 𝑃 ) ⟩ ⟨(5 𝑃 ) ⟩ ⟨( 𝑃3 ) ⟩ ⟨(4 𝑃 ) ⟩
𝐵𝑀6 𝑠2 ; 0.9 𝑠 , −0.1 ; 0.9 𝑠3 ; 1 𝑠 , 0.3 ; 0.8 𝑠 , −0.1 ; 1 𝑠 , −0.4 ; 0.9 𝑠 , 0.4 ; 1
⟨ 𝑃 ⟩ ⟨( 𝑃4 ) ⟩ ⟨( 𝑃 ) ⟩ ⟨( 𝑃3 ) ⟩ ⟨( 3𝑃 ) ⟩ ⟨( 2𝑃 ) ⟩ ⟨( 𝑃4 ) ⟩
𝐵𝑀7 𝑠 ; 0.9 𝑠 , 0.4 ; 0.8 𝑠 , 0.2 ; 1 𝑠 , −0.1 ; 0.9 𝑠 , 0.1 ; 1 𝑠 , −0.2 ; 1 𝑠 , 0.4 ; 0.9
⟨(4 𝑃 ) ⟩ ⟨( 5𝑃 ) ⟩ ⟨( 𝑃5 ) ⟩ ⟨( 3𝑃 ) ⟩ ⟨( 2𝑃 ) ⟩ ⟨( 1𝑃 ) ⟩ ⟨( 5𝑃 ) ⟩
𝐵𝑀8 𝑠4 , 0.2 ; 1 𝑠5 , 0.4 ; 0.9 𝑠5 , 0.2 ; 1 𝑠3 , −0.1 ; 0.9 𝑠2 , 0.1 ; 1 𝑠1 , −0.2 ; 1 𝑠5 , 0.4 ; 0.9

𝐴𝐶8 𝐴𝐶9 𝐴𝐶10 𝐴𝐶11 𝐴𝐶12 𝐴𝐶13


⟨( 𝑃 ) ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩
𝐵𝑀1 𝑠 , 0.3 ; 0.9 𝑠 ; 0.9 𝑠 ; 0.9 𝑠 ; 0.8 𝑠 ; 0.8 𝑠 ;1
⟨ 𝑃4 ⟩ ⟨(2 𝑃 ) ⟩ ⟨(2 𝑃 ) ⟩ ⟨ 4𝑃 ⟩ ⟨(5 𝑃 ) ⟩ ⟨(5 𝑃 ) ⟩
𝐵𝑀2 𝑠 ; 0.9 𝑠 , 0.4 ; 0.9 𝑠 , −0.4 ; 1 𝑠 ; 0.8 𝑠 , −0.3 ; 1 𝑠 , −0.2 ; 1
⟨ 4𝑃 ⟩ ⟨( 3𝑃 ) ⟩ ⟨( 3𝑃 ) ⟩ ⟨ 𝑃4 ⟩ ⟨( 𝑃3 ) ⟩ ⟨( 𝑃4 ) ⟩
𝐵𝑀3 𝑠 ; 0.9 𝑠 , 0.4 ; 0.9 𝑠 , 0.4 ; 1 𝑠 ; 0.9 𝑠 , −0.3 ; 1 𝑠 , −0.1 ; 0.9
⟨ 𝑃4 ⟩ ⟨ 𝑃2 ⟩ ⟨ 𝑃2 ⟩ ⟨ 6𝑃 ⟩ ⟨ 𝑃3 ⟩ ⟨( 3𝑃 ) ⟩
𝐵𝑀4 𝑠 ;1 𝑠 ;1 𝑠0 ; 1 𝑠 ;1 𝑠6 ; 1 𝑠 , −0.2 ; 1
⟨ 6𝑃 ⟩ ⟨(0 𝑃 ) ⟩ ⟨( 𝑃 ) ⟩ ⟨ 6𝑃 ⟩ ⟨( 𝑃 ) ⟩ ⟨( 5𝑃 ) ⟩
𝐵𝑀5 𝑠 ;1 𝑠 , −0.2 ; 0.6 𝑠 , −0.2 ; 0.9 𝑠 ;1 𝑠 , 0.2 ; 0.9 𝑠 , 0.4 ; 1
⟨ 6𝑃 ⟩ ⟨( 1𝑃 ) ⟩ ⟨( 1𝑃 ) ⟩ ⟨(6 𝑃 ) ⟩ ⟨( 5𝑃 ) ⟩ ⟨( 4𝑃 ) ⟩
𝐵𝑀6 𝑠 ; 0.8 𝑠 , −0.4 ; 1 𝑠 , −0.4 ; 1 𝑠 , 0.4 ; 1 𝑠 , 0.4 ; 1 𝑠 , −0.2 ; 0.9
⟨(4 𝑃 ) ⟩ ⟨( 4𝑃 ) ⟩ ⟨( 4𝑃 ) ⟩ ⟨ 𝑃2 ⟩ ⟨ 𝑃2 ⟩ ⟨( 3𝑃 ) ⟩
𝐵𝑀7 𝑠5 , −0.2 ; 0.9 𝑠 , −0.4 ; 1 𝑠 , −0.4 ; 1 𝑠1 ; 0.9 𝑠3 ; 1 𝑠3 , 0.4 ; 1
⟨( 𝑃 ) ⟩ ⟨( 1𝑃 ) ⟩ ⟨( 1𝑃 ) ⟩ ⟨ 𝑃 ⟩ ⟨ 𝑃 ⟩ ⟨( 𝑃 ) ⟩
𝐵𝑀8 𝑠6 , −0.2 ; 1 𝑠1 , −0.4 ; 1 𝑠1 , −0.4 ; 1 𝑠1 ; 0.9 𝑠3 ; 1 𝑠5 , 0.4 ; 1

5.5.3. Comparative analysis between the inclusion and exclusion of QFD- the modified IRs of ACs given by designers to characterize the basic
based analysis tool requirements of office floorings can be deemed as the CIRs of ACs, or
the weights of ACs. Completing the implementation of the proposed
A QFD-based analysis tool has been introduced in this paper to
MCGDM model obtains a revised ranking of office floorings of 𝐵𝑀4 ≻
facilitate the precise identification of CRs and help to transfer CRs into
𝐵𝑀1 ≻ 𝐵𝑀5 ≻ 𝐵𝑀3 ≻ 𝐵𝑀8 ≻ 𝐵𝑀7 ≻ 𝐵𝑀2 ≻ 𝐵𝑀6 , which is
ACs for enhancing client satisfaction level. If we exclude CRs during
quite different from the original ranking of alternatives. Comparing
the proposed MCGDM and assume that the ACs remain the same, then

802
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Table 10
Credibility matrix.
𝐵𝑀1 𝐵𝑀2 𝐵𝑀3 𝐵𝑀4 𝐵𝑀5 𝐵𝑀6 𝐵𝑀7 𝐵𝑀8 𝜑+ 𝜑− 𝜑
𝐵𝑀1 1.000 0.933 0.000 0.000 0.000 0.867 0.237 0.071 3.108 2.018 1.091
𝐵𝑀2 0.407 1.000 0.000 0.000 0.000 0.996 0.000 0.000 2.403 2.420 −0.017
𝐵𝑀3 0.610 0.420 1.000 0.000 0.000 0.000 0.000 0.000 2.030 1.000 1.030
𝐵𝑀4 0.000 0.000 0.000 1.000 1.000 0.839 0.000 0.000 2.839 1.881 0.959
𝐵𝑀5 0.000 0.000 0.000 0.881 1.000 0.875 0.000 0.000 2.755 2.000 0.755
𝐵𝑀6 0.000 0.067 0.000 0.000 0.000 1.000 0.000 0.000 1.067 6.277 −5.210
𝐵𝑀7 0.000 0.000 0.000 0.000 0.000 0.892 1.000 0.833 2.725 2.177 0.548
𝐵𝑀8 0.000 0.000 0.000 0.000 0.000 0.809 0.941 1.000 2.749 1.904 0.845

Fig. 4. Sensitivity analysis.

Fig. 5. Rankings of alternatives obtained by different MCDM methods.

the novel ranking with the original one, we note that the rankings of properties have outstanding performance in terms of basic standards
solid hardwood flooring (𝐵𝑀4 ) and engineered wood flooring (𝐵𝑀5 ) or needs and aesthetics and get higher rankings even if they have poor
have improved, whereas those of rubber flooring tiles (𝐵𝑀1 ) and carpet performance in acoustic properties (𝐴𝐶1 ). On the other hand, from the
tiles (𝐵𝑀3 ) have dropped. The main reason for these changes is that perspective of clients, 𝐵𝑀1 and 𝐵𝑀3 with outstanding performance in
designers usually pay more attention to basic standards or needs and acoustic properties (𝐴𝐶1 ) are preferable.
assign more weights to the relevant criteria, such as hardness (𝐴𝐶2 ),
wear resistance (𝐴𝐶3 ), and impact resistance (𝐴𝐶4 ). In contrast, clients
are more concerned with quietness, comfort and sustainability and so 5.5.4. Comparisons among several well-known ranking methods
the corresponding criteria get higher weights (CIRs), such as acoustic To further demonstrate the necessity of the introduction of ELEC-
properties (𝐴𝐶1 ) and aesthetic appeal (𝐴𝐶8 ). Thus, from the perspec- TRE III in the context of SBMS, this subsection describes a thorough
tive of designers, 𝐵𝑀4 and 𝐵𝑀5 with poor performance in acoustic comparative analysis performed by applying commonly used MCDM

803
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Table 11
Rankings of alternatives obtained by different MCDM methods.
MCDM models Ranking of alternatives
The proposed method (ELECTRE III embedded) 𝐵𝑀1 ≻ 𝐵𝑀3 ≻ 𝐵𝑀4 ≻ 𝐵𝑀8 ≻ 𝐵𝑀5 ≻ 𝐵𝑀7 ≻ 𝐵𝑀2 ≻ 𝐵𝑀6
SAW 𝐵𝑀4 ≻ 𝐵𝑀5 ≻ 𝐵𝑀8 ≻ 𝐵𝑀7 ≻ 𝐵𝑀3 ≻ 𝐵𝑀1 ≻ 𝐵𝑀2 ≻ 𝐵𝑀6
TOPSIS 𝐵𝑀4 ≻ 𝐵𝑀5 ≻ 𝐵𝑀8 ≻ 𝐵𝑀3 ≻ 𝐵𝑀7 ≻ 𝐵𝑀1 ≻ 𝐵𝑀2 ≻ 𝐵𝑀6
VIKOR 𝐵𝑀4 ≻ 𝐵𝑀5 ≻ 𝐵𝑀8 ≻ 𝐵𝑀7 ≻ 𝐵𝑀1 ≻ 𝐵𝑀2 ≻ 𝐵𝑀3 ≻ 𝐵𝑀6
COPRAS 𝐵𝑀4 ≻ 𝐵𝑀5 ≻ 𝐵𝑀7 ≻ 𝐵𝑀8 ≻ 𝐵𝑀3 ≻ 𝐵𝑀1 ≻ 𝐵𝑀2 ≻ 𝐵𝑀6
MULTIMOORA 𝐵𝑀4 ≻ 𝐵𝑀5 ≻ 𝐵𝑀8 ≻ 𝐵𝑀7 ≻ 𝐵𝑀3 ≻ 𝐵𝑀1 ≻ 𝐵𝑀2 ≻ 𝐵𝑀6

methods for material selection, including Simple Additive Weight- We further conduct the comparative analysis of the sensitivity level
ing (SAW) (Chen, 2012; Mousavi-Nasab and Sotoudeh-Anvari, 2018b; using the proposed method (ELECTRE III) against those using SAW,
Wang et al., 2016), TOPSIS (Chen, 2012; Mousavi-Nasab and Sotoudeh- TOPSIS, VIKOR, COPRAS and MULTIMOORA. Likewise, a sensitivity
Anvari, 2018b; Opricovic and Tzeng, 2004; Sánchez-Lozano et al., analysis is also performed to determine the sensitivity level of the
2016; Wang et al., 2016), VIKOR (Gul et al., 2016; Opricovic and Tzeng, ranking of alternatives on the changing CIRs of ACs. For each MCDM
2004), COPRAS (Mousavi-Nasab and Sotoudeh-Anvari, 2018b), and method, we both increase and decrease the CIR of each AC by 30% and
MULTIMOORA (Hafezalkotob et al., 2016; Zavadskas et al., 2017). Be- 60%, respectively, over the 52 experiments concerning the selection
fore conducting the comparative analysis, we ensured that the original of office flooring. The analysis results of ELECTRE III are depicted
assessment information including comprehensive IRs of ACs 𝑨𝑪𝑰 and in Fig. 4(b) while the analysis results of SAW, TOPSIS, VIKOR, CO-
( )
the collective decision matrix 𝑩 = 𝑏𝑚𝑛 8×13 remained the same. This PRAS, and MULTIMOORA are shown in Fig. 6(a) to (e), in that order.
( )
meant that these MCDM methods are performed on the 𝑩 = 𝑏𝑚𝑛 8×13 Evidently, the sensitivity level of those using ELECTRE III is higher
and 𝑨𝑪𝑰, respectively. The SAW method is the simplest and most than those using SAW, TOPSIS, VIKOR, COPRAS, and MULTIMOORA.
commonly used MCDM method, in which the overall performance The main reason for this is the limited compensation ability of ELEC-
of an alternative office flooring can be determined via the weighted TRE III. To avoid the compensation among performances of criteria,
aggregation of its performances in all the criteria using the BULIWA ELECTRE III introduces the indifference threshold and the preference
operator and serves as the basis for ranking alternative office floorings. threshold on each criterion. Meanwhile, to model the veto mechanism
However, before synthesizing, cost criteria had to be converted into in real decision-making scenarios, ELECTRE III also introduces the veto
benefit criteria. TOPSIS is therefore introduced to rank alternative threshold in each criterion. These thresholds are used to determine the
office floorings by determining the distance of alternatives from the concordance indices and discordance indices which fall into the unit
positive ideal solution and the negative ideal solution (NIS), indicating interval, decreasing the range of performance values of alternatives
that the optimal alternative should have the shortest distance from in each criterion and further leading to higher sensitivity of results
PIS and the longest distance from the NIS. VIKOR shares parts of the obtained from the changing CIRs of ACs.
calculation process with TOPSIS. However, it is introduced basically From the findings of the comparative analyses conducted earlier, it
to determine a compromise solution, providing a maximum ‘‘group is convenient to conclude that ELECTRE III can fit very well with prac-
utility’’ for the ‘‘majority’’ and a minimum of an individual regret for tical SBMS problems owing to the outstanding advantages in handling
the ‘‘opponent’’ (Opricovic and Tzeng, 2004). COPRAS is an MCDM the heterogeneity and non-compensation among ACs. While applying
method which works like aggregation operator. Nonetheless, it does ELECTRE III to SBMS, we should realize that the sensitivity level
not need the conversion from cost criteria to benefit criteria. MULTI- using ELECTRE III is greater than the sensitivity level using other
MOORA (Brauers and Zavadskas, 2010) is an MCDM method developed MCDM methods, as it requires the accurate identification of indif-
as an advancement of the MOORA (Brauers and Zavadskas, 2006). It ference threshold, preference threshold, and veto threshold in each
consists of three parts: ratio system, reference point approach, and full criterion.
multiplicative form.

The rankings of alternatives obtained by the proposed method


6. Conclusions
(ELECTRE III), SAW, TOPSIS, VIKOR, COPRAS, and MULTIMOORA
are shown in Table 11 and Fig. 5. It is worth noting that there are
significant differences between the results obtained by the proposed SBMS is a difficult MCGDM task in the wake of the complexity and
method and those by the five MCDM methods. The main reason for uncertainty of practical decision-making contexts, the heterogeneity
the differences is that ELECTRE III is non-compensate while the other and non-compensation of criteria, and the involvement of various types
MCDM methods are compensative. As for ELECTRE III, the outrank- of stakeholders. To help deal with the complexity of SBMS and fill the
( )
ing relation 𝐵𝑀𝑚1 𝐵𝑀𝑚2 𝑚1 , 𝑚2 = 1, 2, … , 𝑀; 𝑚1 ≠ 𝑚2 means that a gap left by the existing literature, this paper has developed a general
sufficient majority of criteria were in favor of this assertion. Further, decision support framework and has put forward a hybrid MCGDM
none of the criteria are strongly against this assertion owing to the model under the uncertain context in accordance with this framework.
non-compensation of the performance among criteria. As for the other In the model, BULI was employed to deal with uncertainty, and an
five MCDM methods, 𝐵𝑀𝑚1 𝐵𝑀𝑚2 indicates that the comprehensive algorithm for CRP within the BULI context was designed to handle
performance of the alternative 𝐵𝑀𝑚1 outranks that of the alternative the consensus reaching issue among experts. Further, while BULI-based
𝐵𝑀𝑚2 , in which the comprehensive performance of an alternative QFD was used to incorporate CRs into SBMS, BULI-based ELECTRE
could be obtained by aggregating its performances in the entire set of III was applied to determine the final ranking of alternatives. Finally,
ACs. For example, the comprehensive performances of 𝐵𝑀4 and 𝐵𝑀5 an empirical application of office flooring selection was presented.
are higher than those of 𝐵𝑀1 under SAW, TOPSIS, VIKOR, COPRAS, Sensitivity analyses conducted in this context revealed that the IRs of
MULTIMOORA. This indicates that 𝐵𝑀4 and 𝐵𝑀5 performed better CRs or the CIRs of ACs occupy a fairly prominent position in SBMS.
than 𝐵𝑀1 overall. Nevertheless, 𝐵𝑀1 performs better than 𝐵𝑀4 and Likewise, a comparative analysis with the existing methods further
𝐵𝑀5 while using ELECTRE III. The main reason for this is that the gap verified the feasibility and rationality of the proposed hybrid MCGDM.
in 𝐴𝐶1 between the performance of 𝐵𝑀1 and the performance of 𝐵𝑀4 It may, therefore, be concluded that the proposed hybrid MCGDM not
or 𝐵𝑀5 is greater than the value of the veto threshold. Meanwhile, the only furnished a rational and effective decision support tool on SBMS
performances of 𝐵𝑀4 or 𝐵𝑀5 in 𝐴𝐶1 cannot be compensated by the for practitioners but also pointed to a new way that researchers can
good performances in other criteria. follow to deal with SBMS.

804
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Fig. 6. Comparisons with other MCDM methods in terms of sensitivity analysis.

The main contributions of the paper can be summarized as distance and similarity measures, and aggregation operators to
follows: manipulate BULI.
• The model introduced CRP to tap into the diverse knowledge
• The proposed model incorporated CRs into SBMS using QFD to and experiences of various stakeholders and further proposes an
ensure that the selected SBM fulfills the requirements of clients. algorithm for CRP under BULI to deal with the consensus reaching
• The paper proposed a novel linguistic approach for modeling and issue.
processing assessment information with high degree of uncer- • The model introduced ELECTRE III to deal with the heterogeneity
tainty, i.e., BULI. Furthermore, it has defined comparison laws, and non-compensation among criteria.

805
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

In practical building design scenarios, in addition to the CRs and the Chen, Z.S., Chin, K.S., Martínez, L., L., Tsui K., 2018b. Customizing semantics for
basic technical requirements, the apt selection of a building material individuals with attitudinal HFLTS possibility distributions. IEEE Trans. Fuzzy Syst.
6 (26), 3452–3466.
is also affected by the choice of other building materials along with
Chen, Z.S., Chin, K.S., Mu, N.Y., Xiong, S.H., Chang, J.P., Yang, Y., 2017. Generating
the building material under selection, which forms a complete building HFLTS possibility distribution with an embedded assessing attitude. Inform. Sci.
assemble. In our future research, effort will be devoted to the selection 394–395, 141–166.
of sustainable building assemblies under uncertainty. Chen, Z.S., Martínez, L., Chin, K.S., L., Tsui K., 2018a. Two-stage aggregation paradigm
for HFLTS possibility distributions: A hierarchical clustering perspective. Expert
Syst. Appl. 104, 43–66.
Acknowledgments
Chen, Z.S., Yang, Y., Wang, X.J., Chin, K.S., Tsui, K.L., 2019. Fostering linguis-
tic decision-making under uncertainty: A proportional interval type-2 hesitant
This work was supported by the National Natural Science Foun- fuzzy TOPSIS approach based on hamacher aggregation operators and andness
dation of China (Grant Nos. 71801175 and 71871171), the Theme- optimization models. Inform. Sci. 500, 229–258.
based Research Projects of the Research Grants Council (Grant no. Colletti, T.A., Renshaw, J.T., Schaefer, R.E., 1998. Stain resistance in resilient sheet
flooring applications: The role of the plasticizer. J. Vinyl Addit. Technol. 4 (4),
T32-101/15-R) and ‘‘the Fundamental Research Funds for the Central
233–239.
Universities’’ (Grant no. 2042018kf0006), the Spanish Government Dikmen, I., Birgonul, M.T., Kiziltas, S., 2005. Strategic use of quality function
Projects (Grant No. TIN2015-66524-P and PGC2018-099402-B-I00), deployment (QFD) in the construction industry. Build. Environ. 40 (2), 245–255.
and partly supported by the Chongqing Technology and Business Uni- Dong, Y.C., Liu, Y.T., Liang, H.M., Chiclana, F., Herrera-Viedma, E., 2018. Strategic
versity Scientific Research Startup Foundation (Grant No. 950318025), weight manipulation in multiple attribute decision making. Omega 75, 154–164.
the National Key R&D Program of China (Grant No. 2018YFC0810600), Eagleton, T., 1990. The Ideology of the Aesthetic. Duke University Process, Durham
and London.
and the Civil Aviation Flight University of China Scientific Research Ferreira, L., Borenstein, D., Santi, E., 2016. Hybrid fuzzy MADM ranking procedure for
Foundation (Grant No. J2018-12). better alternative discrimination. Eng. Appl. Artif. Intell. 50, 71–82.
Figueira, J.R., Greco, S., Roy, B., Slowinski, R., 2013. An overview of ELECTRE methods
References and their recent extensions. J. MultiCriteria Decis. Anal. 20, 61–85.
Franzoni, E., 2011. Materials selection for green buildings: which tools for engineers
Abeysundara, U.G.Y., Babel, S., Gheewala, S., 2009. A matrix in life cycle perspective and architects?. Procedia Eng. 21, 883–890.
for selecting sustainable materials for buildings in Sri Lanka. Build. Environ. 44 Govindan, K., Jepsen, M.B., 2016. ELECTRE: A comprehensive literature review on
(5), 997–1004. methodologies and applications. European J. Oper. Res. 250, 1–29.
Adorno, T.W., 1997. Aesthetic Theory. The Athlone Process, London. Govindan, K., Shankar, K.M., Kannan, D., 2016. Sustainable material selection for
Akadiri, P.O., Chinyio, E.A., Olomolaiye, P.O., 2012. Design of a sustainable building: construction industry A hybrid multi criteria decision making approach. Renew.
a conceptual framework for implementing sustainability in the building sector. Sustain. Energy Rev. 55, 1274–1288.
Buildings 2 (2), 126–152. Greco, S., Figueira, J., Ehrgott, M., 2016. Multiple Criteria Decision Analysis. Springer,
Akadiri, P.O., Olomolaiye, P.O., 2012. Development of sustainable assessment criteria New York.
for building materials selection. Eng. Constr. Archit. Manag. 19 (6), 666–687. Gu, H.M., Wang, S.Q., Neimsuwan, T., Wang, S.G., 2005. Comparison study of thickness
Akadiri, P.O., Olomolaiye, P.O., Chinyio, E.A., 2013. Multi-criteria evaluation model swell performance of commercial oriented strandboard flooring products. For. Prod.
for the selection of sustainable materials for building projects. Autom. Constr. 30 J. 121 (2), 139–147.
(2), 113–125. Gul, M., Celik, E., Aydin, N., Gumus, A.T., Guneri, A.F., 2016. A state of the art
Akao, Y., Mazur, G.H., 2003. The leading edge in QFD: past, present and future. Int. literature review of VIKOR and its fuzzy extensions on applications. Appl. Soft
J. Qual. Reliab. Manag. 20 (1), 20–35. Comput. 46, 60–89.
Arrigoni, A., Pelosato, R., Meliá, P., Ruggieri, G., Sabbadini, S., Dotelli, G., 2017. Life Hafezalkotob, A., Hafezalkotob, A., Sayadi, M.K., 2016. Extension of MULTIMOORA
cycle assessment of natural building materials: The role of carbonation, mixture method with interval numbers: An application in materials selection. Appl. Math.
components and transport in the environmental impacts of hempcrete blocks. J. Model. 40 (2), 1372–1386.
Cleaner Prod. 149, 1051–1061. Häkkinen, T., Belloni, K., 2011. Barriers and drivers for sustainable building. Build.
Berardi, U., 2013. Clarifying the new interpretations of the concept of sustainable Res. Inf. 39 (3), 239–255.
building. Sustainable Cities Soc. 8, 72–78. Harris, G.W., Shaw, S.R., 1988. Slip resistance of floors: users’ opinions, tortus
Bragança, L., Vieira, S.M., Andrade, J.B., 2014. Early stage design decisions: the way instrument readings and roughness measurement. J. Occup. Accid. 9 (4), 287–298.
to achieve sustainable buildings at lower costs. Sci. World J. http://dx.doi.org/10. Harrison, R., Brough, R., 1992. Impact resistance of ceramic tiles and flooring. Int.
1155/2014/365364. Ceram. J.(Testo Stampato) 14 (55), 31–35.
Brauers, W.K.M., Zavadskas, E.K., 2006. The MOORA method and its application to Hashemi, S.S., Hajiagha, S.H.R., Zavadskas, E.K., Mahdiraji, H.A., 2016. Multicrite-
privatization in a transition economy. Control Cybernet. 35 (35), 445–469. ria group decision making with ELECTRE III method based on interval-valued
Brauers, W.K.M., Zavadskas, E.K., 2010. Project management by multimoora as an intuitionistic fuzzy information. Appl. Math. Model. 40 (2), 1554–1564.
instrument for transition economies. Technol. Econ. Dev. Econ. 16 (1), 5–24.
Hatefi, M.A., 2019. Indifference threshold-based attribute ratio analysis: A method for
Bribián, I.Z., Capilla, A.V., Usón, A.A., 2011. Life cycle assessment of building materials:
assigning the weights to the attributes in multiple attribute decision making. Appl.
Comparative analysis of energy and environmental impacts and evaluation of the
Soft Comput. 74, 643–651.
eco-efficiency improvement potential. Build. Environ. 46 (5), 1133–1140.
Herrera, F., Herrera-Viedma, E., Martínez, L., 2008. A fuzzy linguistic methodology to
Brown, S.K., Sim, M.R., Abramson, M.J., Gray, C.N., 1994. Concentrations of volatile
deal with unbalanced linguistic term sets. IEEE Trans. Fuzzy Syst. 16 (2), 354–370.
organic compounds in indoor air-a review. Indoor Air 4 (2), 123–134.
Herrera, F., Martínez, L., 2000. A 2-tuple fuzzy linguistic representation model for
Bunz, K.R., Henze, G.P., Tiller, D.K., 2006. Survey of sustainable building design
computing with words. IEEE Trans. Fuzzy Syst. 8 (6), 746–752.
practices in north America, Europe, and Asia. J. Archit. Eng. 12, 33–62.
Hosking, J., 2017. Wood Hardness Chart. Retrieved 27 May, 2019, from
Bustos, C., Hernández, R.E., Fortin, Y., 2009. Effect of kiln-drying on the hardness and
https://www.hoskinghardwood.com/department/hardwood-floors/Janka-Wood-
machining properties of tamarack wood for flooring. For. Prod. J. 59 (1/2), 71–76.
Hardness-Chart.aspx?dId=7&pageId=12.
Cabeza, L.F., Rincón, L., Vilariño, V., Pérez, G., Castell, A., 2014. Life cycle assessment
(LCA) and life cycle energy analysis (LCEA) of buildings and the building sector: Invidiata, A., Lavagna, M., Ghisi, E., 2018. Selecting design strategies using multi-
A review. Renew. Sustain. Energy Rev. 29, 394–416. criteria decision making to improve the sustainability of buildings. Build. Environ.
Castro-Lacouture, D., Sefair, J.A., Flórez, L., Medaglia, A.L., 2009. Optimization model 139, 58–68.
for the selection of materials using a LEED-based green building rating system in James, D.I., 1985. Slip resistance tests for flooring: two methods compared. Polym.
Colombia. Build. Environ. 44 (6), 1162–1170. Test. 5 (6), 403–425.
Chan, L.K., Wu, M.L., 2002. Quality function deployment: A literature review. European Jeon, J.Y., Ryu, J.K., Jeong, J.H., Tachibana, H., 2006. Review of the impact ball in
J. Oper. Res. 143 (3), 463–497. evaluating floor impact sound. Acta Acust. United ACUSTICA 92 (5), 777–786.
Chavira, D.A.G., Lopez, J.C.L., Noriega, J.J.S., Valenzuela, O.A., Carrillo, P.A.A., 2017. Jin, J., Ji, P., Liu, Y., Lim, S.J., 2015. Translating online customer opinions into
A credit ranking model for a parafinancial company based on the ELECTRE- engineering characteristics in QFD: A probabilistic language analysis approach. Eng.
III method and a multiobjective evolutionary algorithm. Appl. Soft Comput. 60, Appl. Artif. Intell. 41, 115–127.
190–201. Jin, L.S., Mesiar, R., Borkotokey, S., Kalina, M., 2018. Certainty aggregation and the
Chen, T.Y., 2012. Comparative analysis of SAW and TOPSIS based on interval-valued certainty fuzzy measures. Int. J. Intell. Syst. 33 (4), 759–770.
fuzzy sets: Discussions on score functions and weight constraints. Expert Syst. Appl. Kabak, Ö., Ervural, B., 2017. Multiple attribute group decision making: a generic
39 (2), 1848–1861. conceptual framework and a classification scheme. Knowl.-Based Syst. 123, 13–30.
Chen, Z.S., Chin, K.S., Li, Y.L., Yang, Y., 2016. Proportional hesitant fuzzy linguistic Kacprzyk, J., 1986. Group decision making with a fuzzy linguistic majority. Fuzzy Sets
term set for multiple criteria group decision making. Inform. Sci. 357, 61–87. and Systems 18 (2), 105–118.

806
Z.-S. Chen, L. Martínez, J.-P. Chang et al. Engineering Applications of Artificial Intelligence 85 (2019) 783–807

Khoshnava, S.M., Rostami, R., Valipour, A., Ismail, M., Rahmat, A.R., 2018. Rank of Sánchez-Lozano, J.M., García-Cascales, M.S., Lamata, M.T., 2016. Comparative TOPSIS-
green building material criteria based on the three pillars of sustainability using ELECTRE TRI methods for optimal sites for photovoltaic solar farms. Case study in
the hybrid multi criteria decision making method. J. Cleaner Prod. 173, 82–99. Spain. J. Clean. Prod. 127, 387–398.
Kumar, A., Sah, B., Singh, A.R., Deng, Y., He, X.N., Kumar, P., Bansal, R.C., 2017. A Seo, J., Jeon, J., Lee, J.H., Kim, S.M., 2011. Thermal performance analysis according
review of multi criteria decision making (MCDM) towards sustainable renewable to wood flooring structure for energy conservation in radiant floor heating systems.
energy development. Renew. Sustain. Energy Rev. 69, 596–609. Energy Build. 43 (8), 2039–2042.
Labella, Á., Liu, Y., Rodríguez, R.M., Martínez, L., 2018. Analyzing the performance Singh, R.K., Benyoucef, L., 2013. A consensus based group decision making methodol-
of classical consensus models in large scale group decision making: A comparative ogy for strategic selection problems of supply chain coordination. Eng. Appl. Artif.
study. Appl. Soft Comput. 67, 677–690. Intell. 26 (1), 122–134.
Level, 2018. Many building materials can be reused or recycled. Retrieved 5 Stonhard, 2018. What are stain-resistant floors & wall systems? Retrieved 5 June, 2019,
June, 2019, from http://www.level.org.nz/material-use/minimising-waste/reuse- from https://www.stonhard.com/feature-glossary/stain-resistant-flooring/.
and-recycling/. Takano, A., Hughes, M., Winter, S., 2014. A multidisciplinary approach to sustainable
Martínez, L., Herrera, F., 2012. An overview on the 2-tuple linguistic model for building material selection: A case study in a finnish context. Build. Environ. 82,
computing with words in decision making: Extensions, applications and challenges. 526–535.
Inform. Sci. 207, 1–18. Takano, A., Pal, S.K., Kuittinen, M., Alanne, K., Hughes, M., Winter, S., 2015. The effect
Martínez, L., Rodríguez, R.M., Herrera, F., 2015. The 2-Tuple Linguistic Model: of material selection on life cycle energy balance: A case study on a hypothetical
Computing with Words in Decision Making. Springer, New York. building model in Finland. Build. Environ. 89, 192–202.
Martínez, L., Ruan, D., Herrera, F., 2010. Computing with words in decision support Tian, G.D., Zhang, H.H., Feng, Y.X., Wang, D.Q., Peng, Y., Jia, H.F., 2018. Green
systems: an overview on models and applications. Int. J. Comput. Intell. Syst. 3 decoration materials selection under interior environment characteristics: A grey-
(4), 382–395. correlation based hybrid MCDM method. Renew. Sustain. Energy Rev. 81 (1),
Marzouk, M., Azab, S., Metawie, M., 2018. BIM-based approach for optimizing life 682–692.
cycle costs of sustainable buildings. J. Cleaner Prod. 188, 217–226. Torres, R., Salas, R., Bencomo, N., Astudillo, H., 2018. An architecture based on
Melchert, L., 2007. The dutch sustainable building policy: A model for developing computing with words to support runtime reconfiguration decisions of service-based
countries?. Build. Environ. 42 (2), 893–901. systems. Int. J. Comput. Intell. Syst. 11 (1), 272–281.
Mesiar, R., Borkotokey, S., Jin, L.S., Kalina, M., 2018. Aggregation under uncertainty. UN-United Nations, 2015. Transforming our world: The 2030 Agenda for Sustainable
IEEE Trans. Fuzzy Syst. 26 (4), 2475–2478. Development. Resolution adopted by the General Assembly, Seventieth session
Meyer, L., Brischke, C., Welzbacher, C.R., 2011. Dynamic and static hardness of wood: (A/RES /70/1). Retrieved 5 June, 2019, from https://sustainabledevelopment.un.
method development and comparative studies. Int. Wood Prod. J. 2 (1), 5–11. org/post2015/transformingourworld.
Moghimi, V., Jusan, M.B.M., Izadpanahi, P., Mahdinejad, J., 2017. Incorporating user United States Environmental Protection Agency. 2009. Retrieved 5 June, 2019, from the
values into housing design through indirect user participation using MEC-QFD EPA web archive: https://archive.epa.gov/greenbuilding/web/html/whybuild.html.
model. J. Build. Eng. 9, 76–83. Vasto-Terrientes, L.D., Valls, A., Slowinski, R., Zielniewicz, P., 2015. ELECTRE-III-H:
Mousavi-Nasab, S.H., Sotoudeh-Anvari, A., 2018a. A new multi-criteria decision-making An outranking-based decision aiding method for hierarchically structured criteria.
approach for sustainable material selection problem: A critical study on rank Expert Syst. Appl. 42 (11), 4910–4926.
reversal problem. J. Cleaner Prod. 182, 466–484. Wang, P., Zhu, Z.Q., Wang, Y.H., 2016. A novel hybrid MCDM model combining the
Mousavi-Nasab, S.H., Sotoudeh-Anvari, A., 2018b. A comprehensive MCDM-based SAW, TOPSIS and GRA methods based on experimental design. Inform. Sci. 345,
approach using TOPSIS, COPRAS and DEA as an auxiliary tool for material selection 27–45.
problems. Mater. Des. 121, 237–253. WCED, 1987. Our Common Future: A Report from the United Nations World Commis-
Najjar, M., Figueiredo, K., Palumbo, M., Haddad, A., 2017. Integration of BIM and sion on Environment and Development. Oxford University Press, Oxford. Retrieved
LCA: evaluating the environmental impacts of building materials at an early stage 5 June, 2019, from http://www.un-documents.net/our-common-future.pdf.
of designing a typical office building. J. Build. Eng. 14, 115–126. Wiglusz, R., Sitko, E., Nikel, G., Jarnuszkiewicz, I., Igielsand ka, B., 2002. The effect
Nassar, K., Thabet, W., Beliveau, Y., 2003. A procedure for multi-criteria selection of of temperature on the emission of formaldehyde and volatile organic compounds
building assemblies. Autom. Constr. 12 (5), 543–560. (VOCs) from laminate flooring – case study. Build. Environ. 37 (1), 41–44.
Nicoletti, G.M., Notarnicola, B., Tassielli, G., 2002. Comparative life cycle assessment Wood, L.C., Wang, C., Abdul-Rahman, H., Abdul-Nasir, N.S.J., 2016. Green hospital
of flooring materials: ceramic versus marble tiles. J. Cleaner Prod. 10 (3), 283–296. design: integrating quality function deployment and end-user demands. J. Cleaner
Opricovic, S., Tzeng, G.H., 2004. Compromise solution by MCDM methods: A Prod. 112 (1), 903–913.
comparative analysis of VIKOR and TOPSIS. European J. Oper. Res. 156 (2), Wu, Z.B., Xu, J.P., 2016a. Possibility distribution-based approach for MAGDM with
445–455. hesitant fuzzy linguistic information. IEEE Trans. Cybern. 46 (3), 694–705.
Ortiz, O., Castells, F., Sonnemann, G., 2009. Sustainability in the construction industry: Wu, Z.B., Xu, J.P., 2016b. Managing consistency and consensus in group decision
A review of recent developments based on LCA. Constr. Build. Mater. 23 (1), 28–39. making with hesitant fuzzy linguistic preference relations. Omega 65 (3), 28–40.
Palomares, I., Estrella, F.J., Martínez, L., Herrera, F., 2014a. Consensus under a fuzzy Xiong, S.H., Chen, Z.S., Chin, K.S., 2018. A novel MAGDM approach with proportional
context: Taxonomy, analysis framework AFRYCA and experimental case of study. hesitant fuzzy sets. Int. J. Comput. Intell. Syst. 11 (1), 256–271.
Inf. Fusion 20, 252–271. Yan, H.B., Ma, T.J., 2015. A group decision-making approach to uncertain quality
Palomares, I., Martínez, L., Herrera, F., 2014b. A consensus model to detect and manage function deployment based on fuzzy preference relation and fuzzy majority.
non-cooperative behaviors in large scale group decision making. IEEE Trans. Fuzzy European J. Oper. Res. 241 (3), 815–829.
Syst. 22 (3), 516–530. Yuen, K.K.F., 2009a. Enhancement of ELECTRE I using compound linguistic ordinal
Palomares, I., Rodríguez, R.M., Martínez, L., 2013. An attitude-driven web consensus scale and cognitive pairwise comparison. In: 2009 IEEE International Conference
support system for heterogeneous group decision making. Expert Syst. Appl. 40 (1), on Systems, Man and Cybernetics. IEEE, pp. 4864–4869.
139–149. Yuen, K.K.F., 2009b. Enhancement of TOPSIS using compound linguistic ordinal scale
Perratore, E., 2016. Breathe Easier About Your Flooring. Retrieved 5 June, 2019, from and cognitive pairwise comparison. In: 2009 IEEE International Conference on
https://www.consumerreports.org/flooring/breathe-easier-about-your-flooring/. Fuzzy Systems. IEEE, pp. 649–654.
Rodríguez, R.M., Labella, Á, De Tré, G., Martínez, L., 2018. A large scale consensus Yuen, K.K.F., 2014. A hybrid fuzzy quality function deployment framework using
reaching process managing group hesitation. Knowl.-Based Syst. 159, 86–97. cognitive network process and aggregative grading clustering: An application to
Rodríguez, R.M., Labella, A., Martínez, L., 2016. An overview on fuzzy modeling of cloud software product development. Neurocomputing 142, 95–106.
complex linguistic preferences in decision making. Int. J. Comput. Intell. Syst. 9, Zadeh, L.A., 1975. The concept of a linguistic variable and its application to
81–94. approximate reasoning. Inform. Sci. 8 (3), 199–249.
Rodríguez, R.M., Martínez, L., Herrera, F., 2012. Hesitant fuzzy linguistic term sets for Zavadskas, E.K., Bausys, R., Juodagalviene, B., Garnyte-Sapranaviciene, I., 2017. Model
decision making. IEEE Trans. Fuzzy Syst. 20 (1), 109–119. for residential house element and material selection by neutrosophic MULTIMOORA
Sabapathy, A., Maithel, S., 2013. A multi-criteria decision analysis based assessment of method. Eng. Appl. Artif. Intell. 64, 315–324.
walling materials in India. Build. Environ. 64, 107–117.

807

You might also like