You are on page 1of 16

Information Systems 95 (2021) 101639

Contents lists available at ScienceDirect

Information Systems
journal homepage: www.elsevier.com/locate/is

A knowledge-intensive adaptive business process management


framework

Huseyin Kir , Nadia Erdogan
Computer Engineering Department, Computer and Informatics Faculty, Istanbul Technical University, Turkey

article info a b s t r a c t

Article history: Business process management has been the driving force of optimization and operational efficiency for
Received 2 September 2019 companies until now, but the digitalization era we have been experiencing requires businesses to be
Received in revised form 22 August 2020 agile and responsive as well. In order to be a part of this digital transformation, delivering new levels
Accepted 31 August 2020
of automation-fueled agility through digitalization of BPM itself is required. However, the automation
Available online 10 September 2020
of BPM cannot be achieved by solely focusing on process space and classical planning techniques. It
Recommended by Stefanie Rinderie-Ma
requires a holistic approach that also captures the social aspects of the business environment, such as
Keywords: corporate strategies, organization policies, negotiations, and cooperation. For this purpose, we combine
Business process management BPM, knowledge-intensive systems and intelligent agent technologies, and yield one consolidated
Knowledge-intensive processes intelligent business process management framework, namely agileBPM, that governs the entire BPM
Process modeling and execution life-cycle. Accordingly, agileBPM proposes a modeling methodology to semantically capture the
Process adaptation business interests, enterprise environment and process space in accordance with the agent-oriented
Agent-based business process management
software engineering paradigm. The proposed agent-based process execution environment provides
Agile business process management
cognitive capabilities (such as goal-driven planning, norm compliance, knowledge-driven actions, and
dynamic cooperation) on top of the developed business models to support knowledge workers’ multi-
criteria decision making tasks. The context awareness and exception handling capabilities of the
proposed approach have been presented with experimental studies. Through comparative evaluations,
it is shown that agileBPM is the most comprehensive knowledge-intensive process management
solution.
© 2020 Published by Elsevier Ltd.

1. Introduction Companies need to combine advanced automation, cloud com-


puting, computer-powered processes and intelligent algorithms
The three industrial revolutions of the past were all triggered to transform their businesses and stay competitive [2].
by technical innovations: The introduction of steam-powered
1.1. Problem statement
mechanical production at the end of the 18th century, electri-
fication at the beginning of the 20th century and digitalization While computer-powered processes constitute an important
through the introduction of programmable logic controllers for part of the new enterprise transformation, currently businesses
the automation of production processes in the 1970s. Accord- still have very incomplete and deferred control over their pro-
ing to experts, the fourth industrial revolution—triggered by the cess spaces. It is recognized that while classical business pro-
internet—is no longer a ‘‘future trend’’, it is now at the center of cess management (BPM) systems are suitable for predefined and
many companies’ strategic and research agendas [1]. The wave standardized flows, they are less convenient when it comes to
of Industry 4.0 has already started disrupting markets, spawning capturing business processes whose conduct and execution are
new business models, blurring industry boundaries and enabling heavily dependent on an expert’s knowledge-intensive decision-
multi-company virtual business networks and ecosystems. Busi- making tasks [3]. In addition to the traditional needs of efficiency
ness complexity is growing exponentially and current enterprise and optimization, BPM needs to deliver automation-fueled agility
in order to capture knowledge-intensive processes and be a key
systems are becoming insufficient to meet the agility and dy-
driver of digital transformation [4]. New approaches and methods
namism that are required by this new business environment [1].
that extend BPM are needed to enable building complex adaptive
systems that react dynamically to changes and bring order out of
∗ Corresponding author. the chaos.
E-mail addresses: hkir@itu.edu.tr (H. Kir), nerdogan@itu.edu.tr In literature, various researches have been conducted for
(N. Erdogan). decades to ensure the agility of business process management

https://doi.org/10.1016/j.is.2020.101639
0306-4379/© 2020 Published by Elsevier Ltd.
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

systems (BPMS). Some studies have investigated how techniques make more informed and accurate decisions. Knowledge workers
from the planning community could be used to synthesize new are still the main drivers of the business, and next-generation
processes and repair previously defined ones that are no longer BPM systems need to capture and digitalize the behavior of
suitable for a given context [5,6]. Another set of approaches experts in decision-making, cooperation, and negotiation. For this
bringing together ideas from both the Semantic Web and BPM purpose, the agileBPM aims to formalize goal-directed action
communities to automatically mediate between business experts’ selection, process quality assessment, rule compliance control,
requirements and the semantic web services [7–9]. In addition, a exception management, and dynamic cooperation capabilities of
number of studies have investigated goal or precondition and ef- business experts and serves as agent capabilities.
fect based process selection approaches to build adaptive process
management systems [10–12]. 1.3. Paper contribution and organization
While task/service orchestration and planning driven
approaches are considered mature enough for supporting busi-
The agileBPM provides a consolidated methodology that cap-
ness processes, these methodologies show limitations and pit-
tures both cognitive (such as knowledge, rules, goals) and control
falls while dealing with the collaborative, non-deterministic and
flow dimensions of the enterprise process space in an incremen-
evolving nature of the enterprise knowledge-intensive processes
tal manner. The developed multi-agent based process execution
[1]. Besides, these approaches lack to capture the alignment
environment utilizes and infers the captured business process
of planning decisions with corporate strategies and to recog-
knowledge fabric through the whole BPM life-cycle and trans-
nize organization values, beliefs, and policies. As Di Ciccio et al.
[13] stated, knowledge-intensive business process management forms it as the enterprise itself evolves. While most of the existing
should not focus only on a single dimension (such as tasks or approaches [5,6,12] solely focus on the adaptiveness of busi-
flows), it should capture and manage a series of interrelated ness processes through unanticipated exception handling, the
elements (data, actions, rules, goals, processes, knowledge work- proposed agileBPM architecture also enables addressing the prob-
ers and the environment) along all the phases of the process lem from collaboration (enterprise goals, dynamic collaborations,
life-cycle. quality of service) and context awareness (rule adoption and
compliance, data-driven actions, reasoning) dimensions. Accord-
1.2. Solution approach ingly, the agileBPM has been the most comprehensive solution,
providing the majority of the key requirements of knowledge-
The solution introduced in this work for the management of intensive process management systems identified by Di Ciccio
knowledge-intensive business processes with high variability re- et al. [13].
lies on three hypotheses. The first hypothesis is, process design is The remainder of this paper is structured as follows: After
not limited to the modeling of task and control-flows, but rather this introduction, Section 2 presents a consolidated methodol-
data, rule, goal, environment and process perspectives should be ogy for knowledge-intensive business process modeling and in-
captured in a holistic way. Second, to maintain encapsulation troduces a set of models that conceptualize various aspects of
and componentization, process executions should not be man- the business space and enable autonomous runtime agility. Sec-
aged in terms of task-flows altering the enterprise information tion 3 describes the planning module that drives the decision
space, but rather in terms of interacting entities, each with its support through utilizing constructed business models and Sec-
own goals, knowledge, decisions and life-cycle [3]. The ultimate tion 4 presents four use cases that demonstrate the business
goal of BPM systems shifts from maintaining process automation expert assistance capabilities of the proposed approach. Then in
to assisting business expert’s decision making [13]. Accordingly, Section 5, the adaptiveness and exception management capabil-
as the third hypothesis, at least some fragments of the knowl- ities of the agileBPM system are evaluated through simulation.
edge worker’s expertise can be formalized and supported by Section 6 presents the most prominent approaches to business
autonomous proxies. processes automation and accordingly positions the agileBPM
In this sense, the concept of knowledge-based multi-agent solution. The paper concludes with a summary and an outlook
systems, composed of multiple interacting actors is a promising on further research directions.
modeling approach that can simulate the way complex, dynamic
processes emerge from the goal-driven collaboration of individual
2. Knowledge-intensive process modeling methodology
actors at the bottom level [14]. Also, belief–desire–intention (BDI)
agents provide a natural encapsulation for knowledge, goal and
An agent-based business process management system is a
rule driven decision making that knowledge-intensive process
complex adaptive system that is able to autonomously pursue
management requires [11]. The agileBPM utilizes BDI agents as
organizational goals, perceive the business environment, recog-
autonomous entities that pursue business goals, sense physical
and virtual stimuli in the environment, recognize the business nize the application context, and respond accordingly [15]. How-
context, reason about how they should and should not behave, ever, deciding how a given enterprise system should respond
and assist the business expert accordingly. to changes through adopting an agent, knowledge, and business
In modern BPM systems, discrete process design and execu- process-centered view is a challenging task since there are many
tion phases are gradually disappearing and become continuous factors that need to be considered, such as managerial tenden-
and integrated activities of design, execution and adaptation [13]. cies, organizational strategies, infrastructure, legislation, resource
The agileBPM proposes a consolidated modeling methodology availabilities and many more. Given this multitude of factors,
that seamlessly integrates business processes, knowledge- a number of perspectives are often required to understand an
intensive and multi-agent systems modeling paradigms and de- organizational setting. To achieve such capability, the concept
sign components. This integration enables seamless utilization demands a novel framework with normalized reference models
and transformation of design time building blocks (such as that support a multitude of perspectives, formal methods and a
knowledge, rules, goals, processes, events, etc.) through the whole standardized architecture [16]. As mentioned in the literature [1,
process execution phases (planning, execution, decision making 15,17], knowledge-intensive business processes have many no-
and adaptation) driven by the intelligent agents. tions common with the multi-agent concept (such as knowledge,
The objective of this research is to augment, enrich and sup- constraints, goals, actions, events, plans) and are well-suited for
port business activities to make it easier for business experts to being operated by multi-agent systems.
2
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

Fig. 1. Knowledge-intensive process modeling methodology phases. Fig. 2. Sales order strategic relations model.

Yet multi-agent technology is beyond being simply a dis- most fundamental of all design questions: “What is the sys-
tributed software development infrastructure; it is also a mature tem intended for?” The process identification phase of the BPM
software engineering paradigm. In this regard, agent-oriented life-cycle addresses this question by systematically identifying a
business process management [18] is a full life-cycle proposi- collection of interlinked business processes required to achieve
tion, not just a programming or implementation issue. In ad- an organization’s goals [31]. Therefore, when the system goals
dition to providing the required semantic process model and are decomposed and achievement responsibility assigned to the
ontology stack, it is also essential to provide guidance on how BPMS, goals become system requirements [32]. However, un-
these models are applied. Accordingly, seamless integration of se- like classical BPM modeling methodologies, this strategic align-
mantic processes, knowledge-intensive systems and multi-agent ment between the organizational (goals) and operational (pro-
technologies requires a purpose-built method that enables the cess) models is not only a design-time issue but also the driver of
development of common conceptualizations in integrity. autonomous agent behaviors towards organizational goals in the
Through evaluating the most prominent agent-oriented soft- agent-based BPM. Therefore, some AOSE methodologies, such as
ware engineering (AOSE) methodologies [19–25] and process se- Tropos [23] and Prometheus [33], provide means for the goal-
mantization approaches [26–29] in the literature, we propose driven specification of requirements in a machine-processable
an agent-based knowledge-intensive process modeling method- manner and utilize them to make autonomous decisions at run-
ology and complementary toolkit, which ensures that the design time. Similarly, some BPM methodologies propose binding busi-
and development of knowledge and control flow dimensions ness objectives with process models to maintain their co-
of the business processes are carried out in an integrated and evolution [32]. In both AOSE and BPM domain, i* [34] is one
incremental manner within the agent-oriented software engi- of the most commonly used techniques due to its higher ex-
neering life-cycle [30]. Our proposed methodology reuses and pressiveness, actor–goal orientation, and ability to deal with
enhances model definitions from existing BPM and AOSE method- non-functional requirements [23,32]. Accordingly, the agileBPM
ologies to take advantage of their strengths wherever possible. As adopts i* methodology to analyze requirements and understand
shown in Fig. 1, the modeling methodology is structured in five the organizational context.
phases, which span from the business requirements specification
to deployment of constructed models to the agents. In the fol- 2.1.1. Strategic relations model
lowing, we describe and illustrate the five development phases of The first phase of the proposed methodology defines system
the proposed methodology: Requirements Specification, Analysis, requirements via modeling the stakeholders, their objectives, and
Architectural Design, Detailed Design and Deployment. relationships through a strategic relations model. This model tries
The process modeling methodology will be illustrated through to identify why an actor would engage in some tasks, pursue
a sales order process. The process starts with the arrival of a a goal, need a resource, or desire a qualification. The strategic
sales order to the salesperson. The salesperson triggers the supply relations model is represented by a mind map-like directed graph
chain manager (SCM) for the procurement and delivery of order where business actors are modeled as, nodes and the relations
items. Finally, the accounting role executes the relevant payment between actors are represented by edges. These relations may
process and the sales order is completed. In order to maintain the stand for a task to be performed, a resource to be shared, a mutual
simplicity of the case, only relevant parts of the sample processes goal to be achieved, or a non-functional requirement (soft-goal)
will be addressed. to be satisfied. These relations represent system requirements
from each actor’s viewpoint and generate Organizational and
2.1. Requirements specification Business Functions Ontology [35] as an output. Fig. 2 represents
the simplified requirements analysis of a generic sales process
Requirements analysis represents the initial phase in most through defining primary actors sales, accounting and supply
software engineering methodologies that tries to answer the chain management (SCM), and their dependencies.
3
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

and makes processes easily accessible, reusable, and composable


with other processes. Domain models produce the Enterprise
Data Ontology, which defines the schema of the enterprise and
constitutes a part of the agent knowledge bases.

2.3. Architectural design

There are inevitable dependencies and relationships between


process participants and these interactions are captured in the
architectural design phase. While the goal model focuses on
the internal intentional relationships of an actor, the interaction
models enable the modeling orchestration of goals and interac-
tions of the process participants. Both BPM and AOSE research
domains propose similar models for capturing goal-driven in-
teractions of process participants, such as Subject Interaction
Diagrams of subject-oriented BPM [36] and Prometheus agent
Fig. 3. Supply Chain Management (SCM) goal model.
interaction diagrams [22]. For the modeling of these relations,
contrary to a conventional message-oriented interaction mod-
eling approach that focuses on the precise ordering of partic-
2.2. Analysis ular message exchanges, agileBPM adopts the goal based ab-
stractive approach. Goal-oriented interaction models focus on
Business goals and strategies express what an organization the essential nature and purpose of an interaction that empow-
wants to achieve from the business perspective. In order to allow ers autonomous behaviors instead of pushing agents to follow
modeling organizations from an intentional point of view, busi- predefined communicative acts mechanically.
ness goals need to be captured explicitly. Within the scope of the
analysis phase, Strategic Relations models created in the require- 2.3.1. Interaction model
ments specification phase are extended through looking “into” Interaction diagrams are abstract workflows which are not
the actors in order to model internal intentional relationships. fully defined in terms of concrete tasks and messages and are
AgileBPM utilizes goal models that allow a deeper understanding thus not executable. These abstract flows capture the skeleton
of each process participant’s needs and how these requirements of a process, emphasizing the interactions and business goal de-
are met. As one of the key building blocks of knowledge-intensive pendencies in a bird’s-eye view, leaving the concrete definition of
systems, the goal model reveals the operational goals for each the process to the subsequent steps. Interaction models designed
actor by decomposing strategic goals specified in the strategic independently of action and messaging details provide agents
relations model. with great flexibility on runtime decision making and exception
handling.
2.2.1. Goal model Fig. 4 presents an interaction model of the Deliver Order
Fig. 3 represents the goal model of the supply chain man- business scenario that is accomplished through the coordination
agement role. In this model, through each actor’s perspective, of three roles, namely supply chain management, accounting,
strategic goals are decomposed into operational constituents via and customer. In accordance with the interaction model, the
means-ends analysis, contribution analysis, and task/goal decom- role enacting agent determines its actual actions and messages
position methods. Means-ends relations are utilized to specify at runtime to achieve the collaborative goals. Even messaging
alternative ways of achieving a goal. Decomposition relations pro- protocols can change dramatically in accordance with these goal
vide a hierarchical description of intentional elements that make achievement decisions.
up a routine by resolving into subgoals. The contribution relations
provide elaboration of the effects from intentional elements to 2.3.2. Rule model
soft-goals. Goal models developed for each role incrementally Real business scenarios not only have to realize business
form the Enterprise Strategy Ontology. goals and functional requirements but are also influenced by
regulations, rules and business policies. In knowledge-intensive
2.2.2. Domain model systems, business policies are orthogonally applied to all business
The analysis phase ends with the construction of domain scenarios and define conditions which drive the execution of the
models that capture the environment that the agent can sense processes. In addition to studies on applying rules in the context
and process upon. In accordance with Gaia [21], PASSI [25] and O- of knowledge-intensive business processes management [37],
MaSE [20] methodologies, agileBPM utilizes ontologically defined there are also studies on applying rules in the field of multi-
domain models for instituting common vocabularies amongst agent systems; while every agent tries to achieve its individual
the system participants and for defining the resources that are goal, making a multi-agent system act on a common interest is
consumed/produced during the goal achievement process. Do- only possible via organizing agent behaviors and interactions in a
main models describe what kind of entities an agent and process normative way [38–40]. With the unification of approaches both
can deal with and how they relate to each other. In this con- from BPM and AOSE a rule model is constructed.
text, domain models primarily focus on resources identified in In contrast to the approach based on adding conditional
the strategic relations model, such as sales order, delivery info, branches to processes in order to exhibit rule-compliant behav-
and payment receipt. In addition, the use of public ontologies iors, a rule-based approach provides high flexibility through a
is encouraged as a best practice of ontology modeling. While declarative manner. Through the rule model, since rules are not
conventional process models are based on data flows, agileBPM hard-wired into agent behaviors, conditions may change, new
adopts a knowledge-based approach where references of the obligations might be added, existing norms may become obsolete,
knowledge concepts are carried along the process. This enables and explicit representation of semantic constraints can empower
dynamic adaptation to the evolution of the information model more flexible and reasoning-based decision making. In this sense,
4
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

Fig. 4. Deliver order scenario interaction model.

a rule model is composed of regulative rules that ensure compat- if the ‘‘Deliver goods on time’’ soft-goal fails, the re-planning of
ibility with the business policies by being tracked proactively and adaptation behavior is defined as:
reactive rules which are used for optimizing business processes
deliv erOrder(? order) →
and adapting to constraints reactively. ⨂
(a) Regulative Rules: In many enterprise domains, business OSCM deliv erOnTime(? order) OSCM replanGoal(shipping)
processes often have to strictly comply with business policies, ⨂
In this statement, the connective operator associates the
legislations, partner contracts, and semantic constraints, such as
soft-goal and compensation behaviors (such as planning/
‘‘Dangerous goods barred from carriage by IATA regulations shall
replanning a goal, suspending/canceling a behavior).
not be accepted for air carriage’’. These rules are represented
In addition, for many business scenarios, it is unrealistic to
with Semantic Web Rule Language (SWRL) [41] and proactively
assume that all exceptional situations and business conditions
regulate the behavior of their subjects through defining what is can be anticipated at design-time and thus be incorporated into
legal and what is not. From a compliance perspective, regulative the process model a priori. Reactive rules provide high flexibility
rules have deontic effects which are: obligation (O), prohibition through enabling agents to handle these cases at a cognitive level
(Phb) and permission (Prm). without requiring manual process flow updates. ECA rules are
also utilized for dynamically handling exceptions and defining
IATADangerousGoods(? goods) → PhbSCM airShipping(? goods)
actionable rules (such as ‘‘Sales of certain chemicals that exceed
In this statement, the PhbSCM deontic operator represents the a threshold should be approved by the related customs and trade
prohibition of the supply chain management role. The prohibition authority’’) in a declarative manner. Rule models produce the
of A is the obligation of ¬A (the opposite of A) and deontic effects Enterprise Rules and Constraints Ontology which constitutes the
remain in force until it is terminated or fulfilled. foundation of enterprise and agent rule bases.
(b) Reactive Rules: In the legal domain, the term ‘‘rule’’ is
2.4. Detailed design
usually utilized in the regulatory sense. However, rules also apply
as prescriptions that define who (the rule subject) does what
In the detailed design phase, process models are designed
(the action) in which conditions (the norm application case) [42].
based on the interaction models. The design approach of the pro-
Reactive rules enable the event-driven realization of context-
cess models in this study differs from conventional process mod-
dependent actions through monitoring an environmental circum- els at several points. Initially, business processes are addressed as
stance defined by an event, along with a set of conditions and a loosely coupled behavior network of the process participants.
performing the corresponding behaviors with the occurrence of Accordingly, contrary to the swimlane based role distinction of
the observed pattern. The essential syntax of these rules is on conventional approaches, processes are designed from the per-
Event if Condition do Action (ECA) and provides active real-time spective of the relevant roles, and a separate process model is
reactions to events which are ideally suited for detecting and created for each role as in subject-oriented BPM [36]. Secondly,
reacting to events in dynamic computing environments [43]. instead of defining the procedures that should be followed along
Primarily non-functional requirements (soft-goals) and com- the process step by step, flexibility by under-specification is at-
pensation behaviors are represented in this form. For example, tained through goal tasks. The actual binding of goal tasks is
5
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

postponed to the execution phase and empowers the agent to Model is realized, agents retrieve the relevant role packages from
make context-based optimized decisions at runtime. the related organizational authority and each agent updates itself
to include the characteristics of the enacted role. Agents can join
2.4.1. Process model multiple organizations and may exhibit several roles in each of
The process model for the Deliver Order process of the supply them. Through the role enactment, the agent gains a series of
chain management role is presented in Fig. 5, where yellow tasks new goals, a set of soft-goals, rules applied to the role, as well
(Procurement and Shipping) stand for the goal tasks. While not as the base beliefs and domain knowledge required to play the
depicted on the process visual, in accordance with the Business role.
Process Modeling Ontology (BPMO) meta-model [26], a set of In addition, roles determine the enterprise knowledge graphs
ontological metadata are defined in the process model. Firstly, that are accessible to the agent and the actions it can perform.
the process itself and all goal tasks are linked to the related Each time an agent enacts a role, its capabilities (goals) are
goals in the Enterprise Strategy Ontology. Then, all process inputs published through the Directory Facilitator (DF) agent yellow
and outputs are defined in terms of Enterprise Data Ontology page service [45]. In this way, cooperative agents can dynamically
concepts. Finally, functional process and goal task semantics, such discover the supplier agent and collaborate for the achievement
as preconditions and effects, are defined to specify if the agent is of a common goal.
capable of executing the process and process execution concludes
with a desired logical state. The precondition of the Deliver Order 3. Business models in decision support
process is simply ‘‘Sales order should be approved by both customer
and the sales authority’’ and defined as the antecedent (body) of The goal of BPM systems is shifting from maintaining process
a SWRL rule on top of the Enterprise Data Ontology; automation to assisting business experts’ decision making [13]. To
this end, next-generation BPM suites should not only focus on the
approv edBy({salesOrder }, ? salesRole)∧
execution of control flows but also capture the cognitive aspects
approv edBy({salesOrder }, ? customer) of actors and support decision making. The ultimate decision
The effects of the related process is ‘‘order status changed to support that a BPMS can provide is to recommend actions to
completed and the payment date is updated’’ and defined as the achieve strategic goals. For this purpose, the questions to be
consequent (head) of the rule; answered are: which business processes can be utilized to achieve
organizational goals, which process is better, is the actor capable
orderStatus({salesOrder }, ‘‘completed′′ )∧ of executing the process and does process execution cause a rule
issueDate({paymentReceipt }, ? date)∧ violation? In business process management, these questions are
tried to be answered within the scope of process planning. This
paymentDate({salesOrder }, ? date)
section details how business models are used for decision making.
In order to refer to runtime concepts such as process instance In the literature, there are various studies on process plan-
and agent knowledge from the process space, a set of built-ins ning aim to generate a sequence of actions (i.e., plan) from a
({self}, {process} and {paramName}) are provided. These built-in model that defines how actions work, what the world’s initial
references are replaced by the related agent knowledge base state and the goal state is [46]. A significant number of these
entities (respectively the agent itself, the process instance and the studies utilize classical planning approaches for adaptive process
related process parameter) at process instantiation time. planning. However, there is a gap between the research area
The newly created or modified models are registered to the of theoretical (but often unrealistic) planning on the one hand
Process Registry Agent hosting a repository of known process and real-world planning on the other. The theoretical planning
models in BPMO format, so that they can be used in the next steps approach can be very insufficient for enterprise environments
for discovery and enactment. that are non-deterministic due to the unpredictable intentions of
process participants, and where information completeness cannot
2.5. Deployment be guaranteed. Also, in process management, it is more desirable
to adapt a running process through avoiding radically changing
In our framework, agents are active software entities that are the activities already assigned to process participants, and simply
able to enact roles. It is assumed that agents are designed inde- updating parts of the process that need to be adapted and keeping
pendently of any domain or organization and have their own ini- other parts stable [13]. For this purpose, in accordance with
tial (self) beliefs, desires, capabilities and obligations. Agents join agileBPM’s goal based underspecified process models, we adopt
societies by adopting organizational roles in which assignments a goal-driven approach for runtime decision making.
are defined in the agent model. In goal-driven process planning, the concrete activities to be
performed are determined upon the goal task’s enactment on a
2.5.1. Agent model per-case basis [10,47]. For this purpose, agileBPM adopts a best-
In PASSI [25] role identification models, in Gaia [21] agent first search based heuristics approach to recommend the most
models and in Ingenias methodology [24] agent viewpoint mod- promising goal achieving processes that are compatible with the
els are utilized for the assignment of roles to agents. AgileBPM actual context. At the core of this approach lies the selection of
adopts the formalization depicted in Gaia as it is the most influ- ‘‘best’’ actions where the questions to be answered are: what
ential methodology that utilizes roles as an abstraction for system is the best way to constrain the process search space and how
analysis and deploys to agents for execution [44]. to digitalize the process selection heuristics of the knowledge
In role theory, roles are defined as self-conceptions, self- worker for identifying the promising solutions? To this end, ag-
referent cognitions that agents apply to themselves as a conse- ileBPM proposes a three-step solution. First, process discovery is
quence of the social role positions they occupy. Within the scope utilized for efficiently searching and ranking the processes that
of this proposal, a role is identified as a set of cognitive qualities can achieve the given goal with a given parameter space. Second,
R = ⟨G, Gs , σr , πr ⟩, respectively a set of goals (G) that should be process selection is applied to identify the most appropriate
achieved while playing the role, obligations that need to be sat- process through digitalizing quality assessment, organizational
isfied (soft-goals, Gs ), a set of role-related prohibitions (rules, σr ) strategies and business rules’ compliance control heuristics of the
and beliefs of the role (domain knowledge, πr ). Once the Agent knowledge workers. Third, if there is no single process that can
6
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

Fig. 5. SCM role deliver order process model. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this
article.)

achieve the goal, a state-of-the-art external planner is invoked Subsumes match: Request R subsumes process P ⇔ ∀IP ∈
to compose and recommend a solution. In the following, process in(P) ∧ ∃IR ∈ in(R) : IP ⊒ IR ∧ ∀OP ∈ out(P) ∧ ∃OR ∈ out(R) :
discovery, process selection and process composition modules OR ⊒ OP
are introduced, which drive the action recommendation through Subsumed-by match: Request R subsumed by process P ⇔
establishing a context-based planning pipeline. ∀IP ∈ in(P) ∧ ∃IR ∈ in(R) : IP ⊒ IR ∧ ∀OP ∈ out(P) ∧ ∃OR ∈
out(R) : OP ≡ OR ∨ OR ⊑1 OP where ⊒ denotes subsumption
relation and ≡ denotes concept equality in the Domain Model
3.1. Process discovery
taxonomy tree. These semantic relevance degrees are ordered as:
Exact > Plug-in > Subsumes > Subsumed-by.
Adaptations always start with what is currently available for Structural match: Another proposal for evaluating the seman-
use and apply it in new ways that maximize fitness in the im- tic similarity between a pair of concepts is defined by how closely
mediate term. Without knowing what the available options are, they are related in the hierarchy. For this purpose, the taxonomy
there is no way to establish a meaningful heuristic function. tree lowest common ancestor (LCA) distance metric is utilized
In this respect, agents have to be able to autonomously iden- to overcome poorly defined external and internal domain model
tify which behaviors to recommend to adaptively achieve the taxonomies [49].
desired goals. The process discovery module aims at locating Syntactic match: The process matchings identified through
processes that, totally or partially, fulfill a goal from the process logical signature matchings produce high-precision results (see
repository managed by the Process Registry Agent in an efficient Fig. 6) and enable agents to continue their execution without
way. The process discovery encompasses (a) goal model based human intervention. However, most of the domain model ontolo-
goal matching, (b) domain model based semantic and syntactic gies in the semantic web today are disconnected is-a ontologies,
process signature matching (input/output), and (c) the semantic which hamper the effective use of any logic-based matchmaker.
relevance ranking of these processes by the Process Registry The lack of semantic relations (such as owl:equivalentClass and
Agent. In addition to the semantic signatures of the processes, owl:disjointWith etc.) hinders taking full advantage of the ex-
functional process semantics that are encoded in terms of logi- pressiveness of OWL-DL and the implicit knowledge respectively
cal preconditions are also important in specifying the fitness of inferred by a description logic reasoner. In this manner, seman-
a process to current environmental conditions. However, since tic approaches are mostly assisted by syntactic similarity-based
precondition validation requires the process performer agent’s matching approaches [50]. AgileBPM utilizes Jaccard text sim-
conception of the world (knowledge), this operation cannot be ilarity due to its accuracy in this domain [51] and reinforces
handled by the Process Registry Agent itself and is postponed to it with a distributional semantics-based alignment approach.
the process selection phase governed by the performer agent. Word2vec [52] word embeddings trained with enterprise doc-
Semantic process models are defined in BPMO notation with uments are utilized to detect domain model concepts that are
semantic signature parameters encoded in Enterprise Data Ontol- not semantically linked due to inadequate ontology modeling but
ogy concepts and functional semantics (precondition and effects) are likely to relate through evaluating the linguistic similarity
represented as SWRL rule fragments. In this respect, a seman- of the concepts. This approach significantly improves recall by
providing a basic level of alignment assessment across heteroge-
tic process is a 6-tuple P = ⟨id, G, I , O, Pre, Eff ⟩, respectively an
neous ontologies, identifying domain-specific abbreviations and
identifier, goals (G) referring to Goal Model taxonomy, inputs (I)
synonyms. On the other hand, syntactic alignments require data
and outputs (O) referring to resources defined in Domain Model,
mediators at runtime for translating between concepts located in
preconditions (Pre) and effects (Eff ) that defines constraints on
disjoint ontologies. Development of the necessary mediators are
Domain Model concepts. For a process request R and an offer
out of the scope of this framework and it is assumed that the
P, goal matching can be defined as a strict logical subsumption
required data mediators have already been defined and deployed
relation: to a mediator repository [53], brokered by a data mediator agent
Goal match: Process P matches the goal of Request R ⇔ ∀GR ∈ beforehand.
goal(R) ∧ ∃GP ∈ goal(P) : GR ⊒ GP Hybrid match: Each of the methods used for process discovery
Similar to logic-based matching of the semantic web ser- has its strengths and insufficiencies from the accuracy point of
vice signatures [48], the degree of strict logical process signa- view. In order to benefit from the strengths of all approaches,
ture matching is defined as follows. With ⊒1 denoting direct a hybrid solution has been developed that uses these methods
subsumption relation in the Goal Model taxonomy tree. in combination. The hybrid approach first attempts logic-based
Exact match: Process P fully matches with request R ⇔ ∀IP ∈ matching, if cannot find a match, goes for syntactic and structural
in(P) ∧ ∃IR ∈ in(R) : IP ≡ IR ∧ ∀OP ∈ out(P) ∧ ∃OR ∈ out(R) : similarities, respectively.
OP ≡ OR To the best of our knowledge, there is no publicly avail-
Plug-in match: Process P plugs into request R ⇔ ∀IP ∈ able matchmaker or an evaluation dataset for semantic business
in(P) ∧ ∃IR ∈ in(R) : IP ⊒ IR ∧ ∀OP ∈ out(P) ∧ ∃OR ∈ out(R) : processes [54]. For this reason, we have generated a semantic
OR ⊒1 OP business process test repository (namely BPMO-TC4) based on
7
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

Table 2
Process selection criteria BSC.
Perspective Indicator Detailed indicator
Financial Process cost - Direct cost
- Delay cost
Process trend - Frequency
Customer Agility - Easiness of change
Satisfaction - Responsiveness
- Reliability
Internal business Maintainability - Complexity
- Coherence
Learning & growth Objective - Clearness of purpose
- Clearness of responsibilities

3.2. Process selection

The Process Registry Agent provides a relevance ranked list of


goal-directed behaviors to the process performer agent for rec-
ommendation. Yet adaptation is not solely a change in response
to a stimulus, it is a change in the direction of improvement.
Business process management is also described as a “process
optimization process”, hence the quality of service (QoS) is a cen-
tral part of the process selection in achieving high-performance
Fig. 6. Micro-averaged recall/precision process discovery performance.
organizations. The process selection module aims to enhance
intelligent agents with the process quality evaluation heuristics
Table 1
Process discovery and ranking performance.
of business experts. However, this is not a straightforward task.
Firstly, there are differing perceptions of quality corresponding
Hybrid Logic & LCA Logic Jac & w2v Jaccard
to differing perspectives (business and management, information
AUC 0.82 0.79 0.71 0.59 0.57
nDCG 0.851 0.824 0.768 0.628 0.608
technologies, customer, etc.). Secondly, there are no widely ac-
AQRT 3.88 1.04 0.77 3.12 0.55 cepted approaches and quality metrics to assist in this process.
Studies in the literature are mostly incomplete prototypes that
consider only a limited number of factors that are generally
very conceptual [55–58]. One of the most significant strategic
the semantic service discovery standard test collection OWLS- decision-making techniques utilized by business experts to evalu-
TC41 which covers various application domains such as travel, ate the business processes from a strategic management perspec-
tourism and e-health. The generated test collection consists of tive is the Balanced Scorecard (BSC) system [31,59]. In agileBPM,
1083 process offers and 42 test requests with binary and graded based on the works of Cho and Lee [60], a balanced scorecard
(0 to 3) relevance sets where 160 of the processes and 18 of the with quantifiable indicators is utilized to support multi-criteria
requests have preconditions and effects defined in SWRL. Since decision making of the agents (see Table 2).
OWLS does not have goal definitions, evaluation processes do not A balanced scorecard allows decision makers – in this case
agents – to look at the business from four essential perspec-
have goals either, and discovery is driven by logical, structural
tives in order to identify measures that can be used to track
and syntactic matchings for this test set. The generated evaluation
the implementation of a business strategy. In this direction, pri-
dataset is published along with the project source code.2 mary business perspectives and related quality indicators are as
The experimental performance evaluation results of the pro- follows:
cess matchmaker on the generated binary relevance dataset are Financial perspectives on process quality focus on the practical
summarized in Fig. 6. Experiments show that supporting stan- applicability of the process in an economic context. Indicators
dard strict logical signature matching techniques with taxonomy evaluating these aspects focus on cost and trends:
tree lowest common ancestor (LCA) distance metric significantly • Direct Cost: The average operation cost for each process in
improves logic-based matching performance (see Table 1). An terms of the duration of the tasks carried out during the
appropriate combination of logic and non-logic-based hybrid pro- process: the man-month cost of the responsible roles and
cess matching significantly outperforms both types of selection in the cost of the resources consumed during the process.
terms of precision and recall (with 0.82 AUC and 0.8 optimized • Delay Cost: The average additional cost of the process in-
F-score) but at the cost of higher response times. stances that exceed the average execution duration of the
Discovery result ranking performance is evaluated by normal- process.
ized discounted cumulative gain (nDCG) metric for the process • Process Frequency: The number of process instantiations
matching configuration optimized for the highest F-score on the in a fixed time frame which reflects the popularity of the
process.
graded relevance dataset. The discovery, ranking performance
and average query response time (AQRT in seconds) of each Customers’ expectations from enterprises can be summarized as:
approach are summarized in Table 1. Accordingly, the hybrid
• Agility: Customers expect businesses to respond to changes
process discovery approach is adopted by the agileBPM. quickly and this metric aims to measure the adaptability of
processes to these changing conditions. Within the scope of
this project, process agility is achieved through goal tasks.
1 http://projects.semwebcentral.org/projects/owls-tc/ In this respect, the number of goal tasks used in the process
2 https://github.com/asbpm/agilebpm model is utilized as an indicator of process agility.
8
H. Kir and N. Erdogan Information Systems 95 (2021) 101639
{
Imax −Iv alue
• Responsiveness: The responsiveness metric is used to assess Imax −Imin
if Imax − Imin > 0
the process quality by tracking the average cycle time of Iv alue = for − indicators
1 if Imax − Imin = 0
a process to respond to requests from other processes or
clients [31]. Based on the given normalizations, the quality of each discov-
• Reliability: Single momentary process failure can have a ered process is uniformly calculated through the weighted sum
high, or even critical, impact on customers. The reliability in- of quality indicators.
dicator is simply described by the ratio number of successful ∑
executions/number of scheduled executions [61]. Quality(P) = Ivi alue ∗ w eighti
i=1
From an internal business perspective, processes represent orga-
nizational competencies that form the productive capacity of the In order to identify perspective weights, Cho and Lee [60]
organization. As core business assets, processes should be delib- conduct a survey-based approach and identify the customer per-
erately managed; their understandability and maintainability are spective as the most important one with the highest priority
important. (0.42), followed by internal business (0.26), financial (0.17) and
then learning and growth (0.15). However, instead of setting fixed
• Complexity: The complexity level of a business process af- weights, the agileBPM process selection mechanism supports a
fects the time and effort required for its effective learning,
configurable weighting approach for process quality indicators.
maintenance and optimization. In literature, a number of
Process success factors can change over time due to changes in
metrics for evaluating process complexity exists, but only
the environment. For example, in a financial crisis, the priority
a small number of these have been theoretically and exper-
would shift from customer satisfaction to reduction of costs. By
imentally verified [62]. In this context, we use the control-
flow complexity (CFC) metric [63] in agileBPM to evaluate increasing the weight of the ‘‘financial’’ perspective, all process
the simplicity and understandability of the process design. enactments can be prioritized organization-wide in favor of the
∑ cost-efficient options automatically. In addition, each agent au-
CFC (P) = CFCXOR−split (i) + tonomously optimizes its weights based on its conception of the
i∈{XORs of P } world and optimizes its behaviors. For example, in the case of
a failure during the process execution, an agent may waste some
∑ ∑
CFCOR−split (j) + CFCAND−split (k)
resources (e.g., time and budget). Through prioritizing critical per-
j∈{ORs of P } k∈{ANDs of P }
spectives (e.g., cost and reliability) during replanning, the agent
For XOR-splits of the process, the complexity (CFCXOR−split ) may enact another behavior that better fits the current resource
is equal to the number of transitions going out of the split. availabilities.
CFCOR−split corresponds to 2n − 1 where n is the outgoing
transition count and for AND-splits, the complexity is simply
3.3. Process composition
1. The greater values of CFC represents the greater overall
architectural complexity of a process.
• Coherence: A business process model should only comprise Ideally, for a given goal, process discovery and process selec-
those activities and events which are required to achieve tion modules would identify a process compatible with the actual
the goal. Depicting superfluous aspects in a business process execution context from the process repository. In case a process
model decreases understandability. For this purpose, we cannot be found, agileBPM invokes a state-of-the-art external
use the cohesion metric [64] to quantitatively express the planner to synthesize a process that can achieve the objective.
degree to which the elements of a process belong together. For this purpose, the SHOP2 HTN planner is utilized, which de-
compose goals recursively into smaller sub-goals through the
The autonomous process execution environment requires pro-
goal model until it reaches primitive processes and consequently
cesses with clear business targets and responsibility assignments.
recommends a composite process flow [65]. Since agileBPM pro-
• Clearness of purpose: Goals are matched through logical cess model interfaces are fully compliant with WSMO and OWL-S
subsumption and processes with concrete goals should be specifications, BPMO processes can be translated into SHOP2 do-
preferred over abstract ones. The goal’s depth in the goal mains in a sound and complete way. The adoption of a process
hierarchy (DGT) is used for this purpose. The DGT met- composer, in addition to discovery and selection modules, enables
ric measures the length of the longest path from a given automated recommendations of the tasks to be executed next. In
goal G to the root class in the Enterprise Strategy Ontology this way, the knowledge worker is constantly guided to reach the
inheritance hierarchy. goal and the goal is effectively achieved.
• Clearness of responsibilities: This metric aims to ensure that Algorithm 1 summarizes the overall goal-driven process rec-
the assignments of tasks do not cause ambiguity. The use of ommendation procedure of the agileBPM.
concrete roles is encouraged, rather than the abstract roles
that appear in the upper levels of the role hierarchy, and it
is evaluated through the roles’ depth in the role hierarchy. 4. Application scenarios

Quality indicator values need to be normalized to the same scale


to serve a consistent evaluation through diverse measurement Recommending goal-achieving processes for effectively guid-
metrics. For a positive indicator, such as agility, a higher value ing the business expert during the knowledge-intensive process
implies better quality and negative indicators, e.g., complexity, (KIP) enactment is very important. However, as Di Ciccio et al.
display the opposite effect. Accordingly, indicator values are nor- [13] identify, a business process management system should also
malized by the following equations where Imax and Imin are the provide other capabilities (such as data-driven actions, rule man-
maximum and minimum values of the indicator I amongst all agement, and flexible process execution) to support and enable
discovered processes; the knowledge worker adequately. In this section, four potential
{
Iv alue −Imin
business scenarios displaying further knowledge worker assis-
Imax −Imin
if Imax − Imin > 0 tance capabilities of the agileBPM agent-based process execution
Iv alue = for + indicators
1 if Imax − Imin = 0 environment are introduced.
9
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

Algorithm 1 Goal driven process recommendation algorithm While these preconditions may be met during the initial en-
Require: g: goal to enact, πr : knowledge of the active role, σr : role actment of the Deliver Order process, at actual Shipping goal
related rules task execution time, the sensory analysis may reveal that mois-
1: procedure EnactGoal(g) ture level is above limits. The process selection module applies
2: Pcandidates ← ProcessRegistryAgent .discov er(g) ▷ ACL msg precondition rules on the agent knowledge base and checks to
3: while g not achieved nor terminated do ascertain if the rule reasoner infers a violation. In case of the
4: QoSconfig ← calculate QoS weights based on πr precondition failure, the agent informs knowledge worker, skips
5: Sort(Pcandidates , QoSconfig ) that particular Marine Shipping process selection, re-plans the
6: pc ← Pop(Pcandidates ) goal, and recommends another process (such as a Railway Ship-
7: if pc preconditions met in πr then
ping) to the knowledge worker. In this way, the evolution of
8: if pc effects consistent with σr then
the agent knowledge model dynamically influences the actual
9: Recommend(pc )
10: end if process progression.
11: end if Secondly, the agileBPM knowledge-driven process execution
12: end while environment creates an institutional reality and enables agents
13: if recommendation failed for g then to act in an observe–think–act cycle through logical reasoning.
14: Recommend(shop2Compose(g , πr )) These observations can be a specific knowledge state or con-
15: end if ception of an event, and reactive actions can be the initiation
16: end procedure of goal-directed action planning. In contrast to the approach
based on adding conditional branches and exception handlers to
process models at design time in order to control such events, the
knowledge-driven approach provides higher flexibility through
• Scenario 1: Knowledge is a key constituent of KIPs, and not
reacting on a stimulus during process execution without making
only processes modify the enterprise information model,
assumptions about the time these events may occur.
but also the information model should reshape processes.
Knowledge-driven behaviors allow the evolution of the knowl-
Knowledge-driven behaviors scenario captures how pro-
edge to trigger reactive actions and also soft-goals (non-
cess execution is altered and new reactive actions are trig-
functional requirements) to be maintained. For example, ‘‘system
gered through the evolution of the information model.
must maintain a minimum stock amount for material A’’ soft-goal
• Scenario 2: During KIP enactment, unexpected exceptions
can be represented as a reactive rule and trigger a purchase order
and external events may occur, disrupting the current pro-
goal as a compensation action.
cess flow and accordingly require dynamic adaptation of the
KIP. This scenario reveals agileBPM capabilities to deal with stock(materialA) < $min_amount$ →
unanticipated exceptions.
OSCM triggerGoal(purchase(materialA))
• Scenario 3: In a KIP, a knowledge worker must be able to
explicitly define business rules and constraints that influ- Reactive rules are controlled by the agent with each knowl-
ence the process enactment. The definition and propagation edge update operation. In this way, when the agent observes the
of the rules are detailed in the norm adoption and com- rule premises in its knowledge base, it immediately realizes rule
pliance scenario by introducing a new regulation about conclusions and dynamically enacts the compensation goal.
vegetable supply chains.
• Scenario 4: Unlike well-structured business processes, KIPs 4.2. Deal with unanticipated exceptions
are emergent in their nature and may not be fully pre-
specified. The concrete tasks to be performed need to be Real-world business process execution environments are
identified by the knowledge workers’ decisions at runtime. highly dynamic because of diverse, frequently changing situa-
Flexible process execution scenario details agileBPM’s ad tions: shipment delays, production equipment failures, absent
hoc process definition capabilities through an artful SCM participants, change/cancellation of orders. Such changing situa-
quality assurance process. tions lead to deviations from existing plans and fail to achieve
goals unless swift actions are taken to minimize the adverse
4.1. Knowledge-driven behaviors effects [4]. In agileBPM, the reactive rule model is utilized to
recognize exceptional circumstances automatically and to deter-
KIPs are characterized by tasks whose execution is shaped mine the necessary process instance flow adaptations. For this
by the evolution of the knowledge model [13]. In agileBPM, the purpose, failures trigger new obligations, which are the principal
process and knowledge relation are twofold: first, knowledge can motivators for agents to act. Based on obligations, an agent can
shape the process flows through task preconditions defined on dynamically replace/re-plan the failed goal, trigger a repair action,
enterprise data, and secondly, changes in the knowledge model or abort/roll-back the execution.
can trigger goal-seeking processes. In this scenario, an agent selects railway transportation due
As pointed out previously, as the final step of process selection, to its cost-effectiveness for a shipping goal. If the execution of a
functional process semantics encoded in terms of preconditions particular action during goal enactment results in an unexpected
stating how knowledge may constrain the task execution is con- state, such as a strike in railway services, the agent reactively
firmed by the agent. For example, an agent may discover several re-plans the shipping goal or triggers another compensation goal.
shipping options for the ‘‘Shipping’’ goal of the Deliver Order ⨂
process (see Fig. 5). As a first option, a Marine Shipping process OSCM goal(shipping) OSCM replanGoal(shipping)
may require that the actual moisture level of all transferred goods
be within Transportable Moisture Limits (TML) to prevent lique- Exception and other factors emerge a new environment that is
faction that can lead to a cargo shift and accidents. In agileBPM, different from the initial planning state as they alter the context
marine shipping goal’s precondition can be defined as follows: (resource, time, etc.) of the KIP. Therefore, re-planning becomes
necessary in order to effectively reach the desired goal state from
SalesOrder({order }) ∧ hasProduct({order }, ? good)∧ the current state by recommending the most suited task flow. For
moistureLev el(? good, ? ml) ∧ lessThan(? ml, $TML_MAX $) this purpose, the process discovery module runs and discovers
10
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

the same process candidates as before. However, at this time, agent unable to select a norm-consistent process. For example, if
the context-based QoS configuration module analyzes the new every process required for fulfilling one of the agent’s obligations
state, sets up a more convenient process selection configuration, leads to a forbidden state, the agent may not be able to move for-
and initiates a re-planning. Accordingly, the agent selects an- ward and remain norm compliant unless a resolution is applied.
other shipment option, such as airway or highway transportation, In such a case, the agent informs the human expert to resolve the
and recommends to the knowledge worker. In this way, despite conflict or to provide a new norm compliant process.
the unanticipated exceptions, the process goal can be effectively
reached, as the knowledge worker is continuously guided to reach
4.4. Flexible process execution and ad hoc processes
the process goal.

4.3. Norm adoption and compliance Business processes in supply chain management are highly
formal and standardized today. However, there are still many
In order to ensure that a process execution does not result artful processes in the domain that knowledge workers create
in a breach of the rules governing it, agileBPM provides control ‘‘on the fly’’, to cope with many of the situations that arise in
mechanisms for validating and assuring compliance with the their daily work. One of such cases is quality assurance in the
semantic constraints defined in the Rule Model. The use of a perishable products’ supply chain. Challenges include varieties
normative vocabulary and implementation of norms in agent in product types (crop, live-stock, aquaculture, etc.), differenti-
systems have a well-established tradition to improve multi-agent ating quality standards based on destination and product type
coordination and cooperation [39,40]. However, in contrast to (EurepGAP, ISO 22000, IFS, BRC, etc.), varying shelf life and sup-
open societies, adhering with norms and policies is a must in the ply timetables, chemical treatments, warehousing and packaging
enterprise domain to avoid severe sanctions, maintain customer decisions, joint export planning and information sharing strategy
trust, and the prestige of the organization. For that reason, the within SCM networks. Thus, while the outline of these processes
agileBPM framework requires agents to comply with the whole may be stable at an abstract level, the key details are not [4].
set of norms rather than letting agents autonomously decide in An agent-based execution environment that considers intelligent
which to follow. Accordingly, agents adopt norms voluntarily; BDI agents as a proxy for actors in the enterprise ecosystem is
norm government, punishment, and reward mechanisms are out a promising approach that can simulate how processes emerge
of the scope of this work. from the interaction and decisions of autonomous agents at the
Within the scope of this work, norms are utilized as a factor bottom level.
that influences the agent reasoning and deliberation process, nec- In this scenario, the SCM knowledge worker decides to dynam-
essarily in action selection. Since business rules are so diverse and ically run several quality assurance controls based on her domain
variable, it is unlikely that all regulations can be hard-wired into know-how. For this purpose, agileBPM proposes a multi-agent co-
agent behaviors as conditional branches or process preconditions operation based dynamic interaction mechanism that enables the
at design time. The agent must make sure that the effects of its
definition of ad hoc processes at runtime. Ad hoc processes are
actions do not cause violations of the rules. In such cases, process
not dictated ahead of time but emerge as part of the collaboration
effects should be validated against the rule model in advance in
and negotiation between the participants. Multi-agent systems
order to estimate whether process execution concludes with a
utilize contract net protocols (CNP) for the construction and ex-
desired environment state.
ecution of dynamic collaborations, and accordingly, CNPs have
As an example, take a company in the food domain. Let a new
a well-established background in the multi-agent systems [66].
regulation established by some governmental organization define
The agileBPM also supports CNP, where a coordinator agent can
that certain kinds of vegetables must now be sold within a shorter
sell-by date than before. In a conventional BPM system, a business announce a particular goal that is required to be achieved and
expert needs to identify all processes relating to the shipment of other agents in the system (in-house or external) bid for the
the related kind of vegetable and manually change process flows contract to undertake the goal.
accordingly. In the agileBPM system, the new rule is represented Accordingly, the SCM knowledge worker decides to initiate a
as a regulative rule and added to the agent’s rule base: shelf-life test for a specific perishable product in the warehouse.
For this purpose, the coordinator SCM agent queries Directory
SalesOrder(? order) ∧ hasProduct(? order , ? good)∧ Facilitator (DF) agent to discover every agent who can accomplish
GreenVegetable(? good) → PhbSCM (≥ 5 deliv eryTime)(? order) the required goal. DF provides the list of all goal-achieving agents.
Then, the coordinator solicits proposals from potential contractor
The process performer agent discovers the processes that can
agents by issuing a call for proposals that specify the goal, as
fulfill the Shipping goal of the Deliver Order process (such as air,
well as any requirements (deadline, budget, etc.). Each contractor
railway, water, and truck shipping) and sorts the discovered pro-
knowledge worker evaluates the call and rejects or makes a
cesses based on its QoS preferences, such as cost efficiency, and
proposal with some conditions (price, time, etc.). Once the dead-
selects the water shipping process (see Fig. 5). Before executing
the process, the agent realizes the process effects in its virtual line expires, the SCM knowledge worker evaluates the received
knowledge base as if the process has been completed, updates proposals and selects agents to perform the goal. The proposals
the remaining stock and the deliverable’s actual location, and are binding on the participant agents, so once the coordinator
assesses the delivery time through the average process execution accepts the proposal, the participant acquires a commitment to
time. Then the agent checks whether the knowledge base after perform the goal. Once the participant has completed the goal,
the operation leads to a rule conflict. If the average execution time sends a completion message, and results to the coordinator agent.
of the process is higher than the new sell-by date, a regulative Through ad hoc processes, knowledge workers can step by
rule failure occurs. The agent gives up the enactment of the step, identify quality assurance goals to be achieved and related
current shipping approach selection and goes for the next option participants, and negotiations are automatically established. With
that is compliant with the new regulation. the dynamic initiation of several goals (shelf-life test, sensory
While normative specifications enable greater flexibility, they analysis, chemical treatments, etc.), a complex ad hoc quality
also raise a number of challenges. The adoption of new norms assurance process emerges as a result of the collaboration of the
or revision of existing ones may lead to conflicts and render the participants.
11
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

5. Evaluation Table 3
Process fill rates based on various environments and task failure ratios.
Runtime estimate error
To evaluate the adaptiveness and exception management ca-
pabilities of the proposed agileBPM system, a business process 0% 2% 5% 10% 20% 50%

simulator module, which automatically builds business processes HDHB 1 1 1 0.99 0.97 0.86
MDMB 1 1 0.97 0.92 0.76 0.16
and sets up various challenging execution environments, is de-
SDMB 0.98 0.91 0.81 0.65 0.37 0.02
veloped. MDLB 0.95 0.89 0.80 0.61 0.30 0.0
To our knowledge, there are no publicly available real-world
datasets for evaluating semantic business processes in any stan-
dard notation [27,28,54]. In order to evaluate the agileBPM system
on a standard set of business processes, a synthetic workflow • Error rate: In order to evaluate the exception recovery ca-
gallery that is extracted from real-world scientific applications’ pabilities of the system, a failure model is introduced into
execution traces is utilized [67]. The workflow gallery comprises the simulator. The model is characterized by a failure rate
highly parallel processes with real-world execution times of each ranging from 0% to 50%, which defines the probability of
process task. Given customizable configurations are set for the each process task to fail. These failures may stand for var-
simulator; ious environmental stimuli, such as a knowledge conflict,
resource unavailability, or failure of a physical device, that
• Number of tasks in the process is set as 50, and goals from may occur during process execution. The failure time is
the BPMO-TC4 dataset are randomly assigned to each task. randomly determined between the task start and finish time
• Process repository size for goal matching is fixed to 10 pro- and each failure wastes time and budget in accordance with
cesses per goal. The execution time of each process is se- the exception time.
lected randomly by the normal distribution with the mean
equal to the real-world execution time of the related task. Given an execution environment and a fixed task failure prob-
Similarly, process costs are normally distributed with mean ability, the agileBPM simulator generated 5000 process models
equal to the square of the task execution time. with 50 goal tasks randomly picked from the BPMO-TC4 repos-
• Resource availability: In order to observe the behavior of the itory. In total, we tested 120 000 process models. Fig. 7 shows
proposed system under extreme environmental conditions, the constraint satisfaction capabilities of the agileBPM system
various environments with diverse resource (time and bud- under various environment conditions (resource availabilities)
get) availabilities are defined, from tight constraints, where and runtime task error ratios. Box plots show the ratio of cost
only a small number of process ensembles (goal matching to budget, and makespan to deadline of the process ensembles.
combinations) can be completed, to more liberal constraints Whiskers on the plots indicate maximum and minimum values.
where all of the process-matching configurations can be The ratio indicates whether the value (for example, the simulated
completed. The budget constraints are calculated by iden- cost) exceeded the constraint (budget). Values greater than 1
tifying the minimum budget required to execute the min- indicate that the constraint was exceeded. Accordingly, Table 3
imum cost ensemble of each process (MinBudget), and the respectively represents the process fill rates of the same experi-
smallest budget required to execute all possible ensembles ment. A process execution is successful only if both budget and
of the processes in the gallery (MaxBudget): deadline constraints are satisfied.
∑ This set of experiments points out that, in business environ-
MinBudget = min Cost(p) ments with sufficient resources (HDHB), agileBPM can entirely
p∈g handle task failure ratios up to 10% (approx. five task failures
in a process composed of 50 tasks). After task failures, agileBPM

MaxBudget = max Cost(p)
p∈g
dynamically revises its process selection configuration based on
the new resource availabilities and can successfully cope with
This range (MinBudget to MaxBudget) is then divided into exceptions to a certain failure ratio. For a 20% task failure ratio
three equal intervals (lowBudget, midBudget, highBudget) in HDHB environment the agileBPM has planning effectiveness
to determine the budgets to use in each environment. Sim- (process fill rate) of 97% (see Table 3). This means that 4887
ilarly, the deadline constraints are calculated by identifying process instances out of 5000 have successfully completed within
the shortest duration required to execute the ensembles the specified time and budget, while 113 processes failed since
with the shortest critical path of each process (MinDead- they could not be completed within the deadline, as depicted in
line), and by identifying the smallest duration required to Fig. 7. The system even provides satisfactory process completion
execute all possible ensembles of the processes (MaxDead- ratios in 50% failure cases, which indicate a significant system
line), which is the sum of the critical paths of all pro- malfunction and are considered as rare events. The agileBPM
cess ensembles. Accordingly, shortDeadline, midDeadline process selection approach can successfully cope with limited
and longDeadline ranges are identified. resources (66% resource availability for MDMB) and runtime task
failure ratios up to 5%. Additionally, the plots reveal that the

MinDeadline = min criticalPath(p)
system is more successful in dealing with budget constraints,
p∈g
∑ even with a substantial number of errors. As any task failure
MaxDeadline = max criticalPath(p) in the process critical path extends the makespan, such errors
p∈g directly cause the deadline to be exceeded and make it harder
By setting up four environments with separate resource to adapt through re-planning.
availabilities (longDeadline-highBudget (HDHB),
midDeadline-midBudget (MDMB), shortDeadline- 6. Related work
midBudget (SDMB) and midDeadline-lowBudget (MDLB)), it
is ensured that experiments cover the most interesting area For many organizations, process modeling has become an
of the parameter space for the simulation. essential part of documenting, managing and analyzing their
12
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

Fig. 7. Boxplots for deadline and budget ratios under various resource availabilities and task failure rates varying from 0% to 50%.

business operations. Despite its success, there are still some is- The Planlets [6] approach is based on YAWL (Yet Another
sues that BPM systems have to cope with, such as termino- Workflow Language) [71] workflow definition language and, ac-
logical ambiguities and the low degree of automation of the cordingly, provides a more user-friendly process modeling envi-
BPM life-cycle [68–70]. In order to overcome these problems, ronment. While YAWL provides capabilities to capture all sorts
researchers are tackling the construction of process management of flow dependencies between tasks and to handle expected
systems that can manage complex processes, while remaining exceptions on a rule-based basis, Planlets provides dynamic fail-
robust, reactive, and adaptive despite institutional and environ- ure recovery features on top. For this purpose, Planlets extends
mental changes. In the following, we present the most prominent YAWL with modeling of pre- and post-conditions of tasks and
approaches to business processes management automation and the use of planning techniques. In the context of adaptation of
position our proposal based on these studies. In order to sys- workflow instances, each task in the repository is translated into
tematically position agileBPM with regard to existing approaches, the PDDL [72] and in case of a failure, an external planner is
an evaluation framework that identifies the key requirements utilized for synthesizing the needed recovery procedure on-the-
for knowledge-intensive process management systems is uti- fly. As with SmartPM, classical planning imposes some limitations
lized [13]. Evaluation results are presented in Table 4, which for addressing more expressive problems, including incomplete
shows whether a system provides full (+), partial (∼), or no information, contextual preferences and multiple task effects.
support (−) for each requirement. Also, Planlets mainly focuses on the control-flow perspective of
The ADEPT project and its evolutions propose a system with the process, and information modeling is supported in a primitive
adaptive process management and exception handling capabili- way. Moreover, goals and external events are not part of their
ties [12]. For this purpose, the system adopts a structural adapta- specification.
tions approach at both process schema and instance levels. While Also, a number of goal-based approaches [10,11] exist for
this system has the capability to propagate schema changes to maintaining runtime dynamicity. The Go4Flex project combines
related process instances and to alter instance execution flows in BPMN processes with goal-hierarchies for Jadex BDI agents
cases of unanticipated exceptions, all recovery procedures must equipped with BPMN interpreters [73]. For this purpose, they
be defined and triggered manually by a human process designer propose Goal-oriented Process Modeling Notation (GPMN), a spe-
at run-time. It also utilizes a block-structured modeling approach cific language to model goal-oriented business processes [10]. The
through a non-standard custom notation. The ADEPT project pro- system achieves runtime agility by executing associated different
vides a simple learning approach through case-based reasoning. plans for the same goal, depending on the current situation.
Process instance changes are stored as cases in a case base, to Accordingly, the framework captures business context through
be retrieved and reused in order to perform similar changes in a table of typed and named properties defined by the workflow
the future. Based on the ADEPT technology, the AristaFlow BPM engineer. Due to a lack of semantic capabilities, Go4Flex expects
Suite was developed, with the aim of transferring process flexi- processes that intend the same goal to provide the same interface.
bility and adaptation concepts into an industrial-strength process Hence, these systems are only suitable for homogeneous environ-
management system. ments and cannot handle cross-enterprise collaborations. More-
In contrast to ADEPT, SmartPM (Smart Process Management) over, the primitive approach utilized to represent the context is
provides automated adaptation policies that do not require any insufficient to capture the real-world enterprise environment.
human intervention [5]. For this purpose, they utilize situation Di Ciccio et al. [13] state that to provide complete process-
calculus theories to model the domain in which processes are aware system support, the entire set of interrelated requirements
executed, IndiGolog high-level agent programming language to have to be supported. Application of use case scenarios intro-
encode the processes flows, and classical planning to maintain duced in Section 4 compels several requirements. First, actions
the automated adaptation of a process when needed. SmartPM need to be enriched with constraints, such as pre and post-
provides a GUI-based tool that can explicitly represent data and conditions, defined on process data (data-driven actions). Second,
associate it with a process model encoded through the BPMN 2.0 users should be allowed to explicitly define business rules on pro-
notation. However, compared to ontology-based domain mod- cess data (rules and constraints). Third, ad hoc processes should
eling and the knowledge-based decision-making approach pro- be able to dynamically emerge from the collaboration and ne-
posed by the agileBPM, the SmartPM technology stack brings gotiation of process participants (flexible execution). Fourth, the
some restraints. First, situation calculus-based domain modeling system should be able to catch unpredictable exceptions and
is very low level and difficult to understand for the enterprise generate recovery procedures to recover (unanticipated excep-
process management community. Second, the adopted classical tions). Fifth, external events should be able to alter the infor-
planning approach requires a fully observable, static and deter- mation model and, accordingly, adjust the state of the running
ministic domain which is lacking in capturing the real complexity process (model external events). However, Table 4 reveals that
and dynamicity of the enterprises. most of the proposed approaches focus on handling unanticipated
13
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

Table 4
Evaluating some agile process management approaches against the requirements described in Di Ciccio et al. [13].
Requirement agileBPM ADEPT SmartPM Planlets Go4Flex
Data modeling + ∼ + ∼ ∼
Late data modeling ∼ – – – –

Data
Access to data + – – – +
Access to shared data + + + – +
Rule Action
Data-driven actions + ∼ ∼ ∼ ∼
Late actions modeling ∼ + ∼ ∼ –
Rules and constraints + – + – ∼
Late constraints + – – – –
Goal modeling + – ∼ – +
Goal

Late goal modeling + – ∼ – –


Different modeling styles – – + + –
Visibility of knowledge + ∼ – – +
Processes

Flexible execution + + – – –
Unanticipated exceptions + + + + –
Migration of instances – + – + –
Learning from event logs – ∼ – + –
Learning from data – – – – –
Resource/skill modeling + + + + –
Workers’ interaction – – – – –
Knowledge

Workers’ privileges + – – – –
worker

Late resource modeling + – – + –


Late privileges modeling ∼ – – – –
Workers’ decisions + ∼ ∼ ∼ ∼
Model external events + – + + –
Env.

Late modeling of events + – – – –

exceptions. Accordingly, agileBPM is the only proposal that sup- functions, constraints and constitute the initial knowledge of
ports all given use cases by satisfying the majority of the key the BDI agents. Later, agileBPM presents an agent-based pro-
requirements. cess execution environment that enables agents to make au-
On the other hand, since agileBPM is still a prototype system, tonomous intent-steered and knowledge-driven behavior adap-
it lacks production level requirements such as the migration of tations at runtime. Inspired by the decision-making systematic
running instances in case of a process model change. Also, ag- of business experts, the proposed heuristic planning approach
ileBPM has limited support for learning from historical execution aims to reduce the gap between the idealized model of pro-
logs through incorporating average process execution times and cess and physical reality by gradually determining the course
failure ratios in decision making. AgileBPM aims to integrate a of actions as more information becomes available during execu-
process mining toolkit [74], as ADEPT and Planlets, for more tion. Through this infrastructure, process QoS estimation, agent
comprehensive learning capabilities. capabilities evaluation, norm compliance checks, exception han-
dling and knowledge-driven autonomous actions are provided
7. Conclusions that drive the runtime agility.
Superior to existing methods, agileBPM successfully captures
BPM has successfully delivered benefits over a long period social aspects of the business through incorporating actors’ men-
of time and continues to do so, but new realities of business tal states (such as goals, intentions, obligations, prohibitions,
have created new imperatives for business information systems. beliefs, etc.) to multi-criteria decision-making processes. Corre-
Today’s BPM systems need new methods for thinking about in- spondingly, the system utilizes BDI agents as virtual assistants
telligent processes in ways that build on the concepts of col- that pursue business goals, perceive physical and virtual stim-
laboration, adaptiveness, and context awareness. In the litera- uli in the environment, recognize the business context, reason
ture, there already exists a set of research that aim adaptive about how they should and should not behave, and act accord-
processes through implementing business logic with specialized ingly. The empirical study reveals that, in an environment with
agent programming languages and applying classical planning sufficient resources, the agileBPM framework can successfully
techniques on top. However, their adoption by the industry is handle randomly generated runtime disruptions. Compared with
very limited due to their lack of support for industrial standards the existing studies in the literature, the agileBPM system has
and adversities in their implementation. The research activity been the most comprehensive solution, providing most of the
outlined in this study aims to define a general methodology, key requirements of a knowledge-intensive process management
a concrete architecture, and a prototype process management system.
system for automatic adaptation of business processes through With all given features, the agileBPM framework is not only
a knowledge-based approach. For this purpose, we utilize se- a new intelligent process execution infrastructure but also a
mantically enriched BPM standards and ontologically captured complete business process management life-cycle proposal built
business environments along with the techniques from the field on top of existing standards and best practices. It also serves as a
of multi-agent systems. preliminary infrastructure towards enabling the transfer of exten-
This paper proposes an end-to-end agile BPM solution in two sive multi-agent technology research to the knowledge-intensive
stages. First, we propose a knowledge-intensive business pro- business process management domain.
cess modeling methodology, which captures the enterprise and One of the main limitations of the present study is that ag-
business process space in accordance with the existing AOSE ileBPM considers agents to adopt norms voluntarily and fully
paradigms. Incrementally developed reference models semanti- cooperative in nature, which is relatively optimistic for inter-
cally describe the organization, environment, business strategies, organization cooperation and open societies. Secondly, temporal
14
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

constraints [75] are not yet supported, and therefore neither task [21] L. Cernuzzi, A. Molesini, A. Omicini, The gaia methodology process,
preconditions nor effects may refer to time patterns in the cur- in: Handbook on Agent-Oriented Design Processes, Springer, 2014, pp.
141–172.
rent approach. For future work, we plan to extend the agileBPM
[22] J. Khallouf, M. Winikoff, The goal-oriented design of agent systems: a
framework with enhanced electronic social institutions, which refinement of Prometheus and its evaluation, Int. J. Agent-Oriented Softw.
provides assistance to the automated specification of business Eng. 3 (1) (2009) 88–112.
agreements. The development of such institutions is expected to [23] P. Bresciani, A. Perini, P. Giorgini, F. Giunchiglia, J. Mylopoulos, Tropos:
establish a necessary level of trust that enables privately owned An agent-oriented software development methodology, Auton. Agents
Multi-Agent Syst. 8 (3) (2004) 203–236.
agents to safely engage in business interactions by providing [24] J. Pavón, J. Gómez-Sanz, R. Fuentes, Model driven development
contract management, brokering, negotiation mediation and rep- of multi-agent systems, in: European Conference on Model Driven
utation mechanisms that will transform PMS systems into open Architecture-Foundations and Applications, Springer, 2006, pp. 284–298.
enterprise marketplaces. [25] M. Cossentino, From requirements to code with the PASSI methodology,
in: Agent-Oriented Methodologies, Vol. 3690, Idea Group Inc., Hershey, PA,
USA, 2005, pp. 79–106.
Declaration of competing interest [26] C. Pedrinaci, J. Domingue, C. Brelage, T. Van Lessen, D. Karastoyanova,
F. Leymann, Semantic business process management: Scaling up the
The authors declare that they have no known competing finan- management of business processes, in: Semantic Computing, 2008 IEEE
cial interests or personal relationships that could have appeared International Conference on, IEEE, 2008, pp. 546–553.
[27] M. Rospocher, C. Ghidini, L. Serafini, An ontology for the business process
to influence the work reported in this paper.
modelling notation, in: FOIS, 2014, pp. 133–146.
[28] B. Di Martino, A. Esposito, S. Nacchia, S.A. Maisto, Semantic annotation of
References bpmn: current approaches and new methodologies, in: Proceedings of the
17th International Conference on Information Integration and Web-Based
[1] J. Sinur, J. Odell, P. Fingar, P. Harmon, Business Process Management: The Applications & Services, ACM, 2015, p. 14.
Next Wave: Harnessing Complexity with Intelligent Agents, Meghan-Kiffer [29] A. Annane, N. Aussenac-Gilles, M. Kamel, BBO: BPMN 2.0 based ontology
Press Tampa, 2013. for business process representation, in: 20th European Conference on
[2] H. Panetto, M. Zdravkovic, R. Jardim-Goncalves, D. Romero, J. Cecil, I. Knowledge Management, ECKM 2019, 2019, pp. 49–59.
Mezgár, New perspectives for the future interoperable enterprise systems, [30] H. Kir, N. Erdoğan, Agent-based semantic business process manage-
Comput. Ind. 79 (2016) 47–63. ment methodology, in: Advances in Practical Applications of Scalable
[3] M. Dumas, Business process modeling, in: L. Liu, M.T. Özsu (Eds.), Multi-Agent Systems. The PAAMS Collection, Springer, 2016, pp. 145–156.
Encyclopedia of Database Systems, second ed., Springer, 2018. [31] M. Dumas, M.L. Rosa, J. Mendling, H.A. Reijers, Fundamentals of Business
[4] M. Reichert, B. Weber, Enabling Flexibility in Process-Aware Information Process Management, second ed., Springer, 2018.
Systems: Challenges, Methods, Technologies, Springer Science & Business [32] A. Carmo, M. Fantinato, L. Thom, E.P. Prado, M. Spinola, P.C. Hung, An
Media, 2012. Analysis of Strategic Goals and Non-Functional Requirements in Business
[5] A. Marrella, M. Mecella, S. Sardina, Intelligent process adaptation in the Process Management, 2017.
SmartPM system, ACM Trans. Intell. Syst. Technol. (TIST) 8 (2) (2017) 25. [33] Y. Abushark, T. Miller, J. Thangarajah, M. Winikoff, J. Harland, Requirements
[6] A. Marrella, A. Russo, M. Mecella, Planlets: automatically recovering specification via activity diagrams for agent-based systems, Auton. Agents
dynamic processes in YAWL, in: OTM Confederated International Confer- Multi-Agent Syst. 31 (3) (2017) 423–468.
ences‘‘ on the Move to Meaningful Internet Systems’’, Springer, 2012, pp. [34] S.Y. Eric, Social modeling and i*, in: Conceptual Modeling: Foundations and
268–286. Applications, Springer, 2009, pp. 99–121.
[7] K. Kluza, G.J. Nalepa, M. Ślażyński, K. Kutt, E. Kucharska, K. Kaczor, A. [35] W. Abramowicz, A. Filipowska, M. Kaczmarek, T. Kaczmarek, Semantically
Łuszpaj, Overview of selected business process semantization techniques, enhanced business process modeling notation, in: Semantic Technolo-
in: Advances in Business ICT: New Ideas from Ongoing Research, Springer, gies for Business and Information Systems Engineering: Concepts and
2017, pp. 45–64. Applications, IGI Global, 2012, pp. 259–275.
[8] H.H. Hoang, J.J. Jung, C.P. Tran, Ontology-based approaches for cross- [36] A. Fleischmann, W. Schmidt, C. Stary, S. Obermeier, E. Börger, Subject-
enterprise collaboration: a literature review on semantic business process Oriented Business Process Management, Springer Science & Business
management, Enterp. Inf. Syst. 8 (6) (2014) 648–664. Media, 2012.
[9] I. Weber, J. Hoffmann, J. Mendling, J. Nitzsche, Towards a methodology for [37] S. Sadiq, G. Governatori, Managing regulatory compliance in business
semantic business process modeling and configuration, in: International processes, in: Handbook on Business Process Management, Vol. 2, Springer,
Conference on Service-Oriented Computing, Springer, 2007, pp. 176–187. 2015, pp. 265–288.
[10] K. Jander, L. Braubach, A. Pokahr, W. Lamersdorf, K.-J. Wack, Goal-oriented [38] J.F. Hübner, O. Boissier, R. Kitio, A. Ricci, Instrumenting multi-agent organi-
processes with GPMN, Int. J. Artif. Intell. Tools 20 (06) (2011) 1021–1041. sations with organisational artifacts and agents, Auton. Agents Multi-Agent
[11] D. Greenwood, R. Ghizzioli, Goal-oriented autonomic business process Syst. 20 (3) (2010) 369–400.
modelling and execution, in: Multiagent Systems, InTech, 2009. [39] M. Dastani, D. Grossi, J.-J.C. Meyer, N. Tinnemeier, Normative multi-agent
[12] P. Dadam, M. Reichert, The ADEPT project: a decade of research and programs and their logics, in: Knowledge Representation for Agents and
development for robust and flexible process support, Comput. Sci. Res. Multi-Agent Systems, Springer, 2009, pp. 16–31.
Dev. 23 (2) (2009) 81–97. [40] G. Governatori, A. Rotolo, How do agents comply with norms? in: Pro-
[13] C. Di Ciccio, A. Marrella, A. Russo, Knowledge-intensive processes: char- ceedings of the 2009 IEEE/WIC/ACM International Joint Conference on
acteristics, requirements and analysis of contemporary approaches, J. Data Web Intelligence and Intelligent Agent Technology, Vol. 03, IEEE Computer
Semant. 4 (1) (2015) 29–57. Society, 2009, pp. 488–491.
[14] A. Borshchev, A. Filippov, From system dynamics and discrete event to [41] I. Horrocks, P.F. Patel-Schneider, H. Boley, S. Tabet, B. Grosof, M. Dean, et
practical agent based modeling: reasons, techniques, tools, in: Proceedings al., SWRL: A semantic web rule language combining OWL and RuleML, in:
of the 22nd International Conference of the System Dynamics Society, Vol. W3C Member Submission, Vol. 21, No. 79, 2004, pp. 1–31.
22, Citeseer, 2004. [42] G.H. Von Wright, Norm and Action: A Logical Enquiry, Humanities Press,
[15] N.R. Jennings, P. Faratin, M. Johnson, T.J. Norman, P. O’brien, M. Wiegand, 1963.
Agent-based business process management, Int. J. Coop. Inf. Syst. 5 (02n03) [43] A. Paschke, H. Boley, Z. Zhao, K. Teymourian, T. Athan, Reaction RuleML
(1996) 105–130. 1.0: standardized semantic reaction rules, in: International Workshop on
[16] G. Vossen, J. Lechtenbörger, Structuring what you are doing: 20 years of Rules and Rule Markup Languages for the Semantic Web, Springer, 2012,
business process modelling, in: The Art of Structuring, Springer, 2019, pp. pp. 100–119.
227–238. [44] C. Bernon, M. Cossentino, J. Pavón, An overview of current trends in
[17] F. Bergenti, G. Caire, D. Gotta, D. Long, G. Sacchi, Enacting BPM-oriented european aose research, Informatica 29 (4) (2005).
workflows with wade, in: WOA, 2011, pp. 112–116. [45] C. Campo, Directory facilitator and service discovery agent, 2002, FIPA
[18] T.J. Norman, N.R. Jennings, P. Faratin, E. Mamdani, Designing and imple- Document Repository.
menting a multi-agent architecture for business process management, in: [46] A. Marrella, Automated planning for business process management, J. Data
International Workshop on Agent Theories, Architectures, and Languages, Semant. 8 (2) (2019) 79–98.
Springer, 1996, pp. 261–275. [47] B. Burmeister, M. Arnold, F. Copaciu, G. Rimassa, BDI-agents for agile goal-
[19] A. Sturm, O. Shehory, The landscape of agent-oriented methodologies, in: oriented business processes, in: Proceedings of the 7th International Joint
Agent-Oriented Software Engineering, Springer, 2014, pp. 137–154. Conference on Autonomous Agents and Multiagent Systems: Industrial
[20] S.A. DeLoach, J.C. Garcia-Ojeda, The O-MaSE methodology, in: Handbook Track, International Foundation for Autonomous Agents and Multiagent
on Agent-Oriented Design Processes, Springer, 2014, pp. 253–285. Systems, 2008, pp. 37–44.

15
H. Kir and N. Erdogan Information Systems 95 (2021) 101639

[48] M. Paolucci, T. Kawamura, T.R. Payne, K. Sycara, Semantic matching of web [62] G. Polančič, B. Cegnar, Complexity metrics for process models–A systematic
services capabilities, in: International Semantic Web Conference, Springer, literature review, Comput. Stand. Interfaces 51 (2017) 104–117.
2002, pp. 333–347. [63] J. Cardoso, Control-flow complexity measurement of processes and
[49] V. Cross, Y. Wang, Semantic relatedness measures in ontologies using Weyuker’s properties, in: 6th International Enformatika Conference, Vol.
information content and fuzzy set theory, in: The 14th IEEE International 8, 2005, pp. 213–218.
Conference on Fuzzy Systems, 2005, FUZZ’05, IEEE, 2005, pp. 114–119. [64] I. Vanderfeesten, H.A. Reijers, W.M. Van der Aalst, Evaluating workflow
[50] D. Faria, C. Pesquita, E. Santos, M. Palmonari, I.F. Cruz, F.M. Couto, The process designs using cohesion and coupling metrics, Comput. Ind. 59 (5)
agreementmakerlight ontology matching system, in: OTM Confederated (2008) 420–437.
International Conferences‘‘ on the Move to Meaningful Internet Systems’’, [65] E. Sirin, B. Parsia, D. Wu, J. Hendler, D. Nau, HTN planning for web service
Springer, 2013, pp. 527–541. composition using SHOP2, J. Web Semant. 1 (4) (2004) 377–396.
[51] Y. Sun, L. Ma, S. Wang, A comparative evaluation of string similarity [66] FIPA, FIPA contract net interaction protocol specification, 2002, URL: http:
metrics for ontology alignment, J. Inf. Comput. Sci. 12 (3) (2015) 957–964. //www.fipa.org/specs/fipa00029/.
[52] T. Mikolov, I. Sutskever, K. Chen, G.S. Corrado, J. Dean, Distributed repre- [67] R.F. Da Silva, W. Chen, G. Juve, K. Vahi, E. Deelman, Community resources
sentations of words and phrases and their compositionality, in: Advances for enabling research in distributed scientific workflows, in: E-Science (E-
in Neural Information Processing Systems, 2013, pp. 3111–3119. Science), 2014 IEEE 10th International Conference on, Vol. 1, IEEE, 2014,
[53] A. Mocan, E. Cimpian, J. de Bruijn, D13. 3v0. 2 WSMX data mediation, pp. 177–184.
2005, http://www.wsmo.org/TR/d13/d13, 3, v0. [68] F. Lautenbacher, B. Bauer, C. Seitz, Semantic business process modeling-
[54] R. Bergmann, G. Müller, Similarity-based retrieval and automatic adapta- benefits and capability, in: AAAI Spring Symposium: AI Meets Business
tion of semantic workflows, in: Synergies between Knowledge Engineering Rules and Process Management, 2008, pp. 71–76.
and Software Engineering, Springer, 2018, pp. 31–54. [69] M. Hepp, F. Leymann, J. Domingue, A. Wahler, D. Fensel, Semantic business
[55] A. Dikici, O. Turetken, O. Demirors, A case study on measuring process process management: A vision towards using semantic web services for
quality: Lessons learned, in: Software Engineering and Advanced Appli- business process management, in: E-Business Engineering, 2005. ICEBE
cations (SEAA), 2012 38th EUROMICRO Conference on, IEEE, 2012, pp. 2005. IEEE International Conference on, IEEE, 2005, pp. 535–540.
294–297. [70] A. Filipowska, M. Kaczmarek, A. Koschmider, S. Stein, K. Wecel,
[56] F. Heidari, P. Loucopoulos, Quality evaluation framework (QEF): Modeling W. Abramowicz, Social software and semantics for business process
and evaluating quality of business processes, Int. J. Account. Inf. Syst. 15 management-alternative or synergy? J. Syst. Integr. 2 (3) (2011) 54.
(3) (2014) 193–223. [71] A.H. Ter Hofstede, W.M. Van der Aalst, M. Adams, N. Russell, Modern
[57] I. Vanderfeesten, J. Cardoso, J. Mendling, H.A. Reijers, W. van der Aalst, Business Process Automation: YAWL and its Support Environment, Springer
Quality metrics for business process models, in: BPM and Workflow Science & Business Media, 2009.
Handbook, Vol. 144, Future Strategies Inc., Lighthouse Point, FL, USA, 2007, [72] D. McDermott, M. Ghallab, A. Howe, C. Knoblock, A. Ram, M. Veloso, D.
pp. 179–190. Weld, D. Wilkins, PDDL-The Planning Domain Definition Language, Tech-
[58] S. Overhage, D.Q. Birkmeier, S. Schlauderer, Quality marks, metrics, and nical Report CVC TR-98-003/DCS TR-1165, Yale Center for Computational,
measurement procedures for business process models, Bus. Inf. Syst. Eng. 1998.
4 (5) (2012) 229–246. [73] L. Braubach, A. Pokahr, K. Jander, W. Lamersdorf, B. Burmeister, Go4Flex:
[59] R.S. Kaplan, D.P. Norton, Using the balanced scorecard as a strategic Goal-oriented process modelling, in: IDC, Springer, 2010, pp. 77–87.
management system, Harv. Bus. Rev. 85 (7–8) (1996) 150–+. [74] W.M. Van der Aalst, B.F. van Dongen, C.W. Günther, A. Rozinat, E. Verbeek,
[60] C. Cho, S. Lee, A study on process evaluation and selection model for T. Weijters, Prom: The process mining toolkit, BPM (Demos) 489 (31)
business process management, Expert Syst. Appl. 38 (5) (2011) 6339–6350. (2009) 2.
[61] A. Respício, D. Domingos, Reliability of BPMN business processes, Procedia [75] A. Lanz, M. Reichert, B. Weber, Process time patterns: A formal foundation,
Comput. Sci. 64 (2015) 643–650. Inf. Syst. 57 (2016) 38–68.

16

You might also like