Professional Documents
Culture Documents
2022 - Managing Public Opinion in Consensus-Reaching Processes For Large-Scale Group Decision-Making Problems
2022 - Managing Public Opinion in Consensus-Reaching Processes For Large-Scale Group Decision-Making Problems
Guo-Rui Yang, Xueqing Wang, Ru-Xi Ding, Jingjun (David) Xu & Meng-Nan Li
To cite this article: Guo-Rui Yang, Xueqing Wang, Ru-Xi Ding, Jingjun (David) Xu &
Meng-Nan Li (2022): Managing public opinion in consensus-reaching processes for large-
scale group decision-making problems, Journal of the Operational Research Society, DOI:
10.1080/01605682.2021.1993760
ORIGINAL ARTICLE
CONTACT Ru-Xi Ding dingruxi@bit.edu.cn School of Management and Economics, Beijing Institute of Technology, Beijing 10081, PR China
Yangtze Delta Region Academy of Beijing Institute of Technology, Jiaxing.
ß Operational Research Society 2022
2 G.-R. YANG ET AL.
involved (e.g., more than 20 people), has attracted recommendation modification plan, DMs have the
increasing attention (Du et al., 2020; Gou et al., right to reject it (Liu et al., 2019a; Zhang et al.,
2020; Li et al., 2019; Xu et al., 2019). In LGDM 2018), which can have a very large influence on the
events, possible alternatives are evaluated by DMs efficiency of the CRPs used for improving the con-
using a variety of representations with the aim of sensus among DMs.
selecting the most acceptable alternative (Chao Opinion dynamics, which investigates the fusion
et al., 2021a; Lu et al., 2021; Ren et al., 2020). In process of individual opinions, has become a power-
general, LGDM models are designed to rank a ful tool for supporting public opinion management
group of feasible alternatives to reach an acceptable (Dong et al., 2017; Liang et al., 2020). Numerous
consensus among DMs (Li et al., 2020b; Xu et al., opinion dynamics models have been proposed:
2020; Zhang et al., 2018). In addition, the analytic Degroot (1974) proposed an opinion dynamics
hierarchy process is a widely used tool in the prob- model in which people’s opinions in the next stage
lem of multi-criteria group decision making. Lin are the weighted combinations of their opinions in
et al. (2020) proposed aggregation of the nearest the previous stage; Friedkin and Johnsen (1990) pre-
consistent matrices with the acceptable consensus in sented a model to reveal opinion dynamics in com-
analytic hierarchy process-group decision making. plex circumstances; Hegselmann and Krause (2002)
Consensus-reaching processes (CRPs), an import- presented a bounded trust-based opinion dynamics
ant part of LGDM models, have been widely used to model, in which when people update their opinions,
assist DMs by improving the level of consensus only opinions within the bounded trust value will
(Song et al., 2021; Tian et al., 2019). Normally, a be referred to; Durrett et al. (2012) developed a
typical CRP includes four main phases (Ding et al., model in which relationships between people are
2020): (1) the DM clustering phase, (2) the detection updated with the evolution of opinion; Dong et al.
of non-cooperative/key DMs/clusters phase, (3) the (2017) proposed that leadership has an important
preference modification and recommendation-feed- effect on the evolution of opinion, and Li et al.
back phase, and (4) the DM weight-updating phase. (2020a) further presented a model to investigate the
Numerous CRPs have been proposed to deal with change in relationships between in the evolution of
LGDM problems in different situations. For opinion. Moreover, Zha et al. (2020) pointed out
instance, Palomares et al. (2014) presented a CRP to that the application of opinion dynamics model in
detect and the non-cooperative behaviours of DMs. group decision making and e-commerce is begin-
Dong et al. (2016) further proposed an operator ning to be recognized, and it is still necessary to fur-
that can generate the DMs’ weights according to ther develop the theoretical basis for in-depth
their behaviours, and Xu et al. (2019) presented a interdisciplinary integration research.
clustering method for DMs, based on which the Besides, it is worth mentioning that the public
non-cooperative behaviours can be defined and opinion management proposed in this paper refers
managed. Liu et al. (2019b) proposed a trust rela- to matching the evolution of public opinion with
tionship-based CRP to detect and manage DMs’ the development of consensus among DMs, rather
conflict behaviours. than manipulating the public opinion. Traditional
In the recommendation-feedback processes of management measures generally include identifying
CRPs, DMs are often required to provide their eval- false information, making timely clarifications about
uations in pairwise comparison matrices. In Zhang the event, and so on.
et al. (2019), multiple self-confidence levels are con- According to the above literature review, the cur-
sidered in pairwise comparison matrices provided rent research gaps relating CRPs and public opinion
by DMs. Recommendation modification plans, management for public-related LGDM events can be
which are obtained by combining DMs’ individual summarized as follows:
and collective evaluations with a certain modifica-
tion rate (Wu & Xu, 2012), are often provided to Many existing CRPs focus on the clustering
assist DMs in modifying their evaluations. Zhang methods of DMs (e.g., Li et al., 2019), the detec-
et al. (2020) reviewed the feedback mechanism with tion of DMs’ non-cooperative behaviours (e.g.,
minimum adjustment or cost. Chao et al. (2021b) Xu et al., 2019; Gou et al., 2020), and the feed-
proposed a framework to improve the consensus back mechanism with minimum cost (e.g., Zhang
reaching and uses the minimum consensus cost to et al., 2020). As one of the other key factors in
reduce the total cost for LGDM events with more the success of decision-making events, public
than 1000 participants. Several previous studies have opinion is not considered in CRPs, which may
assumed that DMs will accept the recommendation lead to a discrepancy between the consensus
modification plans in the CRPs. However, in real- among DMs and wider public opinion.
world decision-making events, when faced with a Furthermore, without proper management, this
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 3
kind of discrepancy will translate into the failure categories based on their confidence levels which
of LGDM events. is one of the most important factors that can
Most traditional opinion dynamics models are affect the reserving ratio of their previous opin-
designed based on the assumption that people’s ions in the process of updating opinions. General
opinions (o 2 ½1, 1) are in one dimension (e.g., management measures are also divided into two
Degroot, 1974; Dong et al., 2017; Li et al. 2020a). categories for them, which can significantly
However, in a public-related LGDM event, DMs improve the efficiency of the public opinion
are generally required by multi-criteria decision- management process.
making to provide their evaluation
(E 2 RNN , N 2) in multi-dimension on alter- The remainder of this paper is organized as fol-
natives of the decision-making event (e.g., Zhang lows. In Section 2, preliminaries regarding LGDM
et al., 2018; Chao et al., 2021a). In this situation, and opinion dynamics are introduced. Then, two
as one of most important factors in the evolution opinion dynamics models for the public-related
of DMs’ opinions, DMs’ evaluation cannot be LGDM event are presented in Section 3. The POM-
considered in existing opinion dynamics models. CRP to manage public opinion in LGDM events is
Furthermore, the effect of general management described in Section 4. Following this, an illustrative
measures for public opinion is limited. When example and several simulations are provided in
managing public opinion, existing research does Section 5 to demonstrate the validity and efficiency
not distinguish people with their different attrib- of the POM-CRP for managing public opinion.
utes and takes targeted management measures Finally, conclusions are drawn in Section 6.
accordingly. These individual attributes often
have a significant impact on the effectiveness of
management measures. 2. Preliminaries
Before formally introducing the POM-CRP, it is
Motivated by these challenges, this study devel-
necessary to briefly review some preliminaries. A
oped a public opinion management-based consensus
description of LGDM problems is provided in
reaching process (POM-CRP) for public-related
Section 2.1. Additionally, basic concepts of opinion
LGDM events. If the consensus or public opinion of
dynamics are introduced in Section 2.2.
the event cannot reach the acceptable level, the deci-
sion-making event may fail. Thus, the aim of the
POM-CRP is to manage public opinion and 2.1. Basic concepts of public-related
improve consensus among DMs simultaneously, LGDM events
ensuring the smooth process of the event. The inno-
vations of the proposed model can be described Some basic concepts of a public-related LGDM
as follows: event are set out as follows:
In the POM-CRP, public opinion is treated as a 1. Let X ¼ fx1 , x2 , :::, xn , :::, xN g be the alternative
vital indicator of the LGDM event. Before mak- set and f1, 2, :::, n, :::, Ng represent the indexes
ing a final decision, both the consensus among of the alternatives.
DMs and the public opinion need to reach 2. Let D ¼ fd1 , d2 , :::, dm , :::, dM g be the DM set
acceptable levels. The POM-CRP is proposed to and f1, 2, :::, m, :::, Mg represent the indexes of
help DMs reach a consensus while simultan- the DMs.
eously achieving public satisfaction with the 3. Let A ¼ fa1 , a2 , :::, ag , :::, aG g be the public set,
LGDM event, ensuring the smooth implementa- which consists of G people including DMs,
tion of the result of the alternative selec- stakeholders and others who have expressed
tion process. opinions towards the event. Let
In order to investigate the evolution process of f1, 2, :::, g, :::, Gg represent the indexes of peo-
DMs’ opinions, a method is developed to reduce ple, meeting the condition: G M: For con-
the dimension of DMs’ evaluations. Through the venience, in the rest paper, people in set A will
proposed method, DMs’ evaluation on alterna- be called agents.
ij
tives of the decision-making event will be trans- 4. Let Em ¼ ðem ÞNN be DM dm’s evaluation
ij
lated into a one-dimensional value, which is an matrix, where em represents dm’s degree of pref-
important part of DMs’ one-dimen- erence for alternative xi over
sional opinions. xj ði, j 2 f1, 2, :::, n, :::, NgÞ:
To achieve a more effective public opinion man- 5. Let E ¼ ðE1 , E2 , :::, Em , :::, EM Þ be the evaluation
agement process, people are divided into two matrix including all Em :
4 G.-R. YANG ET AL.
6. Let x ¼ ðx1 , x2 , :::, xm , :::, xM Þ be the decision indicates al and ak do not know each other. On
weights of the DMs, where xm represents the these bases, we define several formal concepts of
importance weight of DM dm in the opinion dynamics as:
LGDM event.
1. Let cg ¼ ðog , clg Þ be agent ag’s state vector in
LGDM events, which consists of the individual
2.2. Basic concepts of opinion dynamics
opinion og and the confidence level clg.
In this paper, we use og 2 ½1, 1 to represent agent 2. Let C ¼ ðc1 , c2 , :::, cg , :::, cG Þ be the state vector
ag’s individual opinion of an LGDM event: og ¼ 1 of all agents in LGDM events.
means that ag completely supports the LGDM event; 3. Let R ¼ ðrlk ÞGG ðrlk 2 f0, 1gÞ be the relation-
1 indicates total opposition; and 0 indicates neu- ship matrix of all agents, meeting the condi-
trality. The POM-CRP allows agents to provide their tion: rll ¼ 0 ðl ¼ 1, 2, :::, GÞ:
individual opinions, between total opposition and
total support. Besides, the false information refers to
agents’ opinions o<lðl 2 ð1, 0ÞÞ: The setting of l 3. Opinion dynamics models for
is often related to the targeted public opinion. For LGDM events
example, if the targeted public opinion is 0.25, l Social media provides agents with quick and con-
is often set as l<0:25: Moreover, DMs’ individual venient access to information, and agents may gen-
opinions can be calculated based on their evalua- erate opinions relating to an LGDM event through
tions on alternatives. More details are shown in different social media platforms. As there is often a
Section 3.2.2. discrepancy between the generated public opinion
With the development of the LGDM event, and the consensus reached by DMs, public opinion
every agent will continuously form new individual has become one of the key factors in the success of
opinions, and these will contain a part of his/her public-related LGDM events. Thus, it is necessary to
previous individual opinion. It is worth mention- fully understand the evolution of public opinion and
ing that for different agents, the persistence of carry out appropriate management measures for
their own previous individual opinions will be dif- LGDM events.
ferent. Agents with high confidence levels are In this section, the public opinion calculation
more likely to stick to their opinions. Conversely, process is presented in Section 3.1. Opinion
less confident agents are more likely to be affected dynamics models to describe the evolution of dif-
by others’ opinions. Therefore, we use clg 2 ½0, 1 ferent agents’ opinions in an LGDM event are
to represent agent ag’s confidence level: clg ¼ 1
introduced in Section 3.2. In Section 3.3, agents are
indicates that ag is full of confidence and com-
divided into different categories based on their
pletely unaffected by other agents, while 0 indi-
attributes, and targeted management measures for
cates that ag is not at all confident and there will
these categories with different attributes
be no reservation of his/her previous opinion
are proposed.
when forming a new individual opinion. However,
in the real world, for agents, communicating on
social media is different from face-to-face commu- 3.1. Public opinion calculation process
nication. Social media magnifies the impact of In this part, we present a public opinion calculation
agents’ opinions on others, which means that the
operator to compute the public opinion produced
number of agents whose confidence levels belong
by all agents. Let z ¼ f1, 2, :::, Zmax g be a discrete
to ½0:5, 1 is quite small. Besides, there are few
time index that is also the index of the consensus
agents whose confidence levels are close to 0 in
round. Here, we utilize the simple aggregation of
real life. Therefore, we assume that most agents’
agents’ individual opinions as the public opinion,
confidence levels belong to ½0:1, 0:5:
which is defined below.
Besides, we use the directed approach to repre-
sent the relationships between agents, and the Definition 3.1. (Public Opinion) For each consen-
relationships are assumed to be bidirectional. Every sus round of POM-CRP, let
pair of agents al and ak can choose one of four PG z
g¼1 og
ways to represent their relationships: ðrlk , rkl Þ ¼ PO ¼
z
(1)
ð1, 0Þ, ðrlk , rkl Þ ¼ ð0, 1Þ, ðrlk , rkl Þ ¼ ð1, 1Þ, or ðrlk , rkl Þ G
¼ ð0, 0Þ: If rlk ¼ 1, we refer to al as a follower of ak; be the public opinion towards the LGDM event in
(1, 0) and (0, 1) indicate that the relationship the z-th consensus round, where PO 2 ½1, 1 :
between al and ak is unidirectional; (1, 1) indicates PO > 0 indicates that most of the public hold a posi-
that al and ak are followers of each other; and (0, 0) tive attitude towards the LGDM event; conversely, if
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 5
PO < 0, we can infer that the LGDM event is he/she is following. It worths mentioning that when
resisted by most agents. the value of G is large enough, opinions from other
agents he/she is not following can be replaced by
the public opinion, since the public opinion is
3.2. Opinion dynamics
obtained by aggregating all agents’ opinions.
Since public opinion is calculated by aggregating In communication between agents, since agents
agents’ individual opinions, public opinion will are more easily influenced by others they trust, trust
evolve gradually with the evolution of agents’ indi- relationships are also important. Let T ¼ ðtlk ÞGG be
vidual opinions. In this section, a typical opinion the trust matrix among agents, where tlk represents
dynamics model will be introduced first. Then, dif- the trust of agent al toward agent ak. A large value
ferent opinion dynamics models for different agents of tlk indicates that agent al will be more easily
will be proposed. In the end, the evolution process affected by agent ak. If rlk ¼ 0, which means that al
of public opinion is presented. is not following ak, the trust value assigned to ak by
In existing researches, several opinion dynamics al will equal to 0.
models have been proposed (e.g., Degroot 1974; Following the above idea, the individual opinion
Hegselmann & Krause, 2002; Dong et al., 2017; Li of agent ag who does not belong to the DM set in
et al. 2020a). In most previous models, agent ag’s the ðz þ 1Þ-th consensus round, ozþ1g , can be calcu-
new individual opinion is calculated based on his/ lated by:
her previous individual opinion and the opinions of
others within a certain distance from ag in the social ozþ1
g ¼ clg ozg þ ð1clg Þ w1 POz
network. For example, in Dong et al. (2017), agent
ag’s opinion will be updated as a weighted aggrega- X
G
tgl
tion of the opinions of all agents as follows: þ w2 PG rgl ozl
l¼1, al 2D l¼1, al 2D tgl
X
G !
ozþ1
g ¼ bg ozg þ ð1bg Þ ngl ozl , X
G
tgk
l¼1, l6¼g
þ w3 PG rgk ozk , (2)
k¼1, ak 62D k¼1, ak 62D tgk
where bg represents the trust value that agent ag
where clg is the confidence level of agent ag; and
gives to his/her own opinion, and ngl, which is nega-
w1 , w2 , w3 2 ½0, 1 (w1 þ w2 þ w3 ¼ 1) represent
tively related to graph distance between ag and al in
the weight of public opinion, the weight of opinions
social network, represents the extent of trust that ag
of DMs ag is following, and the weight of opinions
gives to agent al.
of other agents who are not in the DM set and ag is
following, respectively.
3.2.1. Opinion dynamics model for agents who
are not DMs
The development of social media has gradually nar- 3.2.2. Opinion dynamics model for DMs
rowed the gaps between agents; agents can easily In CRPs, DMs will be required to provide their evalu-
obtain information posted by others they are follow- ation on the alternatives of the LGDM event. In every
ing. Social media also provide agents with access to consensus round of CRPs, every DM will be provided
opinions towards the same event from other agents with a recommendation modification plan to help all
he/she is not following. Moreover, since DMs have DMs reach an acceptable consensus. Then, faced with
participated in the evaluation process of the deci- the modification plan, a DM can choose to accept or
sion-making event, agents are more likely to be reject it. When a DM chooses to reject it, he/she is more
affected by them. Because of these, an opinion likely to be unsatisfied with the decision-making event.
dynamics model for LGDM events is proposed here. Conversely, when a DM decides to accept the modifica-
During the development of LGDM events, agents tion plan, his/her support degree towards the LGDM
will constantly communicate with each other event is often related to the deviation between his/her
through social media. Therefore, when agent ag evaluation and the collective evaluation of all DMs. Since
forms a new individual opinion ozþ1 g in the the collective evaluation is used to make the final decision
ðz þ 1Þ-th consensus round, he/she is mainly of the event, the bigger the deviation between his/her
affected by two things: his/her previous individual evaluation and the collective evaluation is, the lower his/
opinion towards the LGDM event and opinions pro- her support degree towards the event will be. Following
vided by other agents. The latter can be further div- the above idea, a DM’s evaluation, which is in multi-
ided into three parts according to the sources of the dimension, can be transformed into the support degree,
comments: (1) opinions from other agents he/she is which is a one-dimensional value. Before giving the def-
not following; (2) opinions provided by DMs he/she inition of the support degree, the collective evaluation
is following; and (3) opinions voiced by other agents will be defined first.
6 G.-R. YANG ET AL.
Definition 3.2. (Collective Evaluation) Let Ec ¼ event. It is also worth mentioning that at the begin-
ij g g
ðec ÞNN be the collective evaluation matrix, where ning of an LGDM event, DM dm ’s (Because DM dm
ij ij ij also belongs to agent set, g refers to the index of
eijc ¼ x1 e1 þ x2 e2 þ þ xm eijm þ þ xM eM ,
agents and m indicates the index of DMs.) initial
(3) opinions towards the event can be calculated based
g
xm represent the decision weight of DM dm, and on dm ’s initial evaluation through Eq. (4), which
ij ij
ec , em represent the collective preference degree and means that o0g ¼ /0m : With the development of the
DM dm’s preference degree for alternative xi over xj, LGDM event, other agents’ opinions will be consid-
respectively. ered by DMs when updating their opinions.
Definition 3.3. (Support Degree) For DM dm, we In a public-related LGDM event, DMs, usually
define the support degree /zm towards the LGDM composed of the government and relevant experts,
event in the z-th consensus round as: are often required to listen to stakeholders and others.
g
Based on the above analyses, for DM dm , his/her
8 z
< 2devmc
1 , if DM dm accepted the recommendation modification plan,
/zm ¼ maxl ðdevlcz Þ (4)
:
1, if DM dm rejected the recommendation modification plan:
In the above definition, devzmc reflects the devi- individual opinion in the ðz þ 1Þ-th consensus
ation between the collective evaluation Ezc ¼ ðec ÞNN
ij
round is mainly affected by his/her support degree
and dm’s evaluation Ezm ¼ ðem ÞNN in the z-th con-
ij
in the ðz þ 1Þ-th consensus round calculated by
sensus round. This can be calculated using: Eq. (4), individual opinions of agents following
XN 1 XN him/her, and his/her previous individual opinion.
2
devmc ¼ ðdðeijm , eijc ÞÞ, (5) Following the above idea, the individual opinion
NðN 1Þ i¼1 j¼iþ1 g
of DM dm in the ðz þ 1Þ-th consensus round, ozþ1 g ,
can be computed using:
where !
XG
1
dðeijm , eijc Þ ¼ jpijm pijc j: (6)
g ¼clg og þð1clg Þ w4
ozþ1 rgl ozl þw5 /zþ1
z
PG m ,
l¼1 l¼1 r gl
It is clear that / 2 ½1, 1: For DM dm, a large value
of /zm indicates a high degree of support from DMs (7)
g
in the z-th consensus round. Conversely, if /zm <0, where clg is the confidence level of DM dm ;
/4 ,/5 2
we can infer that dm is not satisfied with the LGDM ½0,1 (/4 þ/5 ¼1) represent the weight of opinions
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 7
g g
of agents following dm and the weight of dm ’s sup- words, the weight of public opinion w1 and the
port degree in the ðzþ1Þ-th consensus round, weight of DMs’ opinions w2 will decrease with the
respectively. It is also worth mentioning that differ- increment of the agent’s confidence level cl.
ent from the opinion dynamics model for agents Conversely, if an agent is not the stakeholder of the
who are not DMs, because DMs are often required event, he/she may be easy to be affected by the
to treat their followers equally, agents whose opin- opinions of others. In this situation, he/she is more
ions are referred by DMs will be assigned equal likely to be influenced by the DMs he/she is follow-
importance weights. ing, especially those who have large numbers of fol-
lowers. Meanwhile, public opinion will also play an
3.2.3. Evolution of public opinion important role in the evolution of his/her opinion.
Applying Eqs. (2) and (7), opinions of all agents can For a DM, if he/she has a high confidence level,
be updated. Since public opinion is an aggregation which means that he/she places more trust in his/
of agents’ opinions, public opinion can also be cal- her own opinion, he/she may refer less to the opin-
culated by using Eq. (1). To make the evolution ions of the agents following him/her and more to
process easier to follow, an example of public opin- his/her support degree / for the LGDM event when
ion evolution is provided in Figure 1. Agent a2, who forming his/her new opinion. Contrarily, if he/she is
is a follower of agent a1 and DM d14 , will form his less confident, more weight will be assigned to the
individual opinion based on his previous individual opinions of agents following him/her in the evolu-
opinion, public opinion, DM d14 ’s individual opin- tion of his/her opinion. Therefore, the weight of
ion, and agent a1’s individual opinion. Then, DM opinions of the agents following him/her w4 will
d14 , who is following by agents a1, a2, and a3, can decrease with the increasing confidence level.
update her individual opinion based on her previous In existing researches, confidence levels are often
individual opinion, support degree towards the regarded as one of the most vital parameters in
event in the z-th consensus round, and the individ- opinion dynamics models (e.g., Hegselmann &
ual opinions of agents a1, a2, and a3. Commonly, by Krause, 2002; Dong et al., 2017; Li et al., 2021). In
implementing different opinion dynamics models, this paper, the confidence level also plays an import-
we can obtain the individual opinions of all DMs ant role in the proposed opinion dynamics models.
and agents, and these can then be used to calculate In order to manage the public opinion, different
the public opinion. management measures have been proposed.
Through simulations, we found that the efficiency
of management measures is greatly affected by the
3.3. Management of public opinion
confidence level of agents. More details are given in
In this section, two common categories of agents Sections 5.2.2 and 5.2.3. To reach a higher effi-
are introduced. Targeted management measures are ciency, agents are divided into two common catego-
then proposed to govern the public opinion formed ries: agents with higher confidence levels and agents
by these different categories. with lower confidence levels.
As one of the most important attributes of For the former pattern, they are less affected by
agents, confidence level often has a decisive effect public opinion and opinions of DMs. However, for
on the reserved ratio of agents’ previous opinions the latter pattern, they are more easily affected by
when they update their opinions, which has been the public opinion and opinions of DMs who have
discussed in Section 2.2. Besides, the confidence large numbers of followers. The average confidence
PG
level of agents can also affect the extent of being clg
influenced by the opinions of different agents, which level of all agents can be calculated as CL ¼ g¼1 G :
means that agent’s confidence level has an influence We then define a confidence-level threshold CL,
on the weights of opinions (e.g., w1, w2, w3, w4 and and if CL>CL, we consider that the group belongs
w5) in opinion dynamics models. to the first pattern; otherwise, we consider that it
For an agent who is not DM, if he/she is highly belongs to the second pattern. Normally, agents’
confident about an LGDM event, which means that confidence levels can be obtained in two ways. The
the trust value he/she gives to his/her previous opin- first is a self-reported confidence level (Liu et al.,
ion is relatively large, he/she is more likely to be the 2019c), which is given by the agent and therefore
stakeholder whose self-interest is closely related to lacks objectivity. The second method requires agents
the result of the event. Under the circumstances, no to provide their opinions in intuitionistic fuzzy rep-
matter the public opinion and the opinion of a DM resentation. In this way, their degrees of uncertainty
who often owns numbers of followers are, he/she can be calculated based on their opinions, which is
will be less affected by them because his/her interest negatively correlated with confidence levels of agents
is bound up with the result of the event. In other (Urena et al., 2015).
8 G.-R. YANG ET AL.
The following management measures are intro- the basis of consensus to guarantee the smooth
duced for the two categories of agents: implementation of the result of the LGDM event.
The framework consists of two parts: (1) the public
Management Measure I: For agents with high
opinion management-based consensus-reaching pro-
confidence levels, the best management measure
cess, which is presented in Section 4.1, and (2) the
is identifying false information, making a timely
selection process, which is introduced in Section 4.2.
clarification about the LGDM event, and expos-
ing the agents who released this information.
Because the interests of highly confident agents 4.1. Public opinion management-based
are often closely related to the result of the deci- consensus-reaching process
sion-making event, when they are faced with In this section, the POM-CRP is described. In con-
false information (The effect of false information trast to previous CRPs, the POM-CRP no longer
on the evolution of public opinion will be regards improving the level of consensus among
detailed discuss in Section 5.2.2), they will be DMs as the only purpose of the CRP but takes both
more cautious about these opinions. Therefore, if public opinion and the degree of group consensus
these agents are given help to recognize the
into consideration. The POM-CRP consists of four
actual situation, they will not only change their
parts: (1) the DM weight-calculation process, (2) the
views on the LGDM event but will also reduce
consensus and public opinion measurement process,
their reference to the false information, which
(3) the recommendation-feedback process, and (4)
can effectively improve the value of public opin-
the evolution and management of public opinion.
ion PO. In the following simulations, we suppose
These four parts will be elaborated in the next four
that applying management measure I, agents
subsections. The flowchart of POM-CRP is shown
who provide false information will have their
in Figure 2.
trust values reduced by other agents. This means
that if agent ag’s individual opinion in the z-th
4.1.1. DM weight-calculation process
consensus round oz1 g <lðz 1Þ, the trust values In this process, a DM weight-calculation operator is
z
assigned to him/her t , g will be reduced to introduced based on the relationship and trust
ð1q1 Þt, g : The deducted part of trust values
z1
g
matrix among agents. Normally, if DM dm has a
q1 t
z1
, g will then be equally divided among the
PG
large value of rg ¼ l¼1 rlg , which means he/she
agents whose individual opinions oz1 l:
has a large number of followers, he/she may be
Management Measure II: For agents who are less
more respected by the rest of agents. Meanwhile, if
confident and more easily affected by the false infor- g P
dm has large value of tg ¼ Gl¼1 tlg , which means
mation, the best management measure is paying
that he/she is more trusted by others, he/she will
more attention to DMs’ opinions and trying to
have a great influence on other agents. Following
improve DMs’ support degree of the LGDM event. g
this, for DM dm , his/her weight in the LGDM event
Since agents with low confidence are easier to be
can be computed by:
affected by others’ opinions, especially DMs’ opinions, PG
when most DMs support the event, the values of l¼1 tlg rlg
xm ¼ PG PG : (8)
these agents’ opinions will be improved accordingly,
k¼1, ak 2D l¼1 tlk rlk
increasing the value of PO with high efficiency. In P
this paper, we assume that through persuasion, DMs It is clear that Mm¼1 xm ¼ 1: If a DM has a large
who accept the recommendation modification plan is number of followers and is trusted by others, he/she
willing to raise their support degree in the z-th con- will have a large weight in the LGDM event.
sensus round by q2, which means after applying man- Conversely, a small weight often means that a DM
agement measure II, /zm ¼ /zm þ q2 (if the adjusted is less popular than others with high weights.
/zm >1, then /zm ¼ 1).
4.1.2. Consensus and public opinion measure-
The evolution and management of public opinion ment process
is a part of the proposed model. In the next section, In this section, a consensus calculation method is
the complete POM-CRP for public-related LGDM presented to obtain the group consensus degree
events will be presented. (GCD) among DMs, which is one of the most
important indicators to measure the consensus
among DMs. Then, this is considered alongside the
4. Public opinion management-based
public opinion computed by the method proposed
consensus-reaching framework
in Section 3.1: if public opinion and the GCD both
In this section, we propose a framework for public- reach preset thresholds, a selection process will be
related LGDM events to guide public opinion on conducted to choose the optimal alternative.
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 9
PM1 PM
k¼lþ1 xlk ð1devlk Þ
Otherwise, the recommendation-feedback process
l¼1
and/or the management of public opinion will GCD ¼ PM1 PM , (9)
be conducted. l¼1 k¼lþ1 xlk
Normally, the consensus calculation method is
designed to compute the GCD among DMs by tak- where xzlk 2 ½0, 1 can be obtained by xzlk ¼
ing DMs’ evaluations and weights into account. minðxzl , xzk Þ, and devlk reflects the deviation
Several calculation methods have been proposed between dl’s and dk’s evaluations Ezl and Ezk in the z-
(e.g., Wu & Xu, 2012; Palomares et al., 2014). Here, th consensus round.
the method proposed by Palomares et al. (2014) will It is clear that GCD 2 ½0, 1: A large value of
be adopted, which is as follows. GCD indicates a high level of consensus among
Definition 4.1. (Group Consensus Degree) For an DMs; conversely, if there is a large deviation among
LGDM event, the GCD is defined as: DMs’ evaluations, a small value of GCD will be
10 G.-R. YANG ET AL.
work, including POM-CRP and the selection pro- By using Eq. (8), the weights of DMs x0 in the ini-
cess, is presented in Appendix A. tial stage can be computed x0 ¼ ð0:0585, 0.0517,
0.0543, 0.0532, 0.0551, 0.0542, 0.0466, 0.0490,
5. Illustrative example and experiments 0.0514, 0.0472, 0.0454, 0.0430, 0.0495, 0.0496,
0.0510, 0.0433, 0.0457, 0.0490, 0.0508, 0:0514Þ:
In this section, an illustrative example is provided in
Section 5.1 to verify the practicality of the proposed 5.1.1.2. consensus and public opinion measure-
model. In Section 5.2, several simulations are ment process. In this part, GCD and PO are com-
conducted to test the efficiency of the model on puted. By applying Eqs. (5) and (9) on E0, the initial
managing public opinion in the consensus reach- group consensus degree is calculated as GCD0 ¼
ing process. 0:6847: Using Eq. (4), the DMs’ initial individual
12 G.-R. YANG ET AL.
opinions can then be computed. Then, utilizing Eq. adjusted /zm >1, then /1m ¼ 1) to ensure that the
(1), the initial public opinion can be calculated as LGDM event can go smoothly. Then, by using Eqs.
PO0 ¼ 0:0127: Now, because GCD0 ¼ 0:6847<1 (2) and (7), opinions of agents except DMs and
and PO0 ¼ 0:0127<2 , the POM-CRP will be exe- opinions of DMs in the first consensus round are
cuted to improve the consensus among DMs and calculated, respectively.
manage public opinion.
5.1.2.3. DM weight-calculation process. For Case
5.1.2. First consensus round A, applying Eq. (8), the weights of DMs x1 in the
For the POM-CRP, a consensus round comprises first consensus round can be calculated x1 ¼
four parts: (1) recommendation-feedback process, ð0:0584, 0.0516, 0.0539, 0.0537, 0.0549, 0.0547,
(2) evolution and management of public opinion, 0.0465, 0.0492, 0.0514, 0.0467, 0.0458, 0.0435,
(3) DM weight-calculation process, and (4) consen- 0.0494, 0.0494, 0.0513, 0.0434, 0.0452, 0.0489,
sus and public opinion measurement process. It is 0.0506, 0:0515Þ:
worth mentioning that, in the evolution and man- For Case B, it is worth mentioning that because
agement of public opinion, different management only Management Measure II is adopted, which
measures are used for different categories of agents. means that the trust relationship will not be
Here, we consider two situations (Case A: the aver- adjusted with the development of the POM-CRP,
age confidence level of agents CL ¼ 0.5 and Case B: the DMs’ weights in Case B will not change. In
the average confidence level of agents CL ¼ 0.1) to other words, x1 ¼ x0 :
verify the availability of targeted management meas-
ures for different categories of agents. Moreover, for 5.1.2.4. Consensus and public opinion measure-
Case A, the efficiency of management measure II is ment process. By applying Eqs. (1) and (9), GCD
relatively low. Therefore, only Management Measure and PO in this consensus round can be obtained as
I will be adopted. Conversely, only Management Case A: GCD1 ¼ 0:7839, PO1 ¼ 0:0306; Case B:
Measure II will be adopted for Case B. The relation- GCD1 ¼ 0:7820, PO1 ¼ 0:1274: It is clear that in
ship between agents’ confidence levels and the effi- both cases, GCD1 <1 and PO1 <2 : Therefore, the
ciency of different management measures will be second consensus round is activated to help DMs
investigated in Sections 5.2.2 and 5.2.3. reach a consensus and to manage the pub-
lic opinion.
5.1.2.1. Recommendation-feedback process. Using Due to the limitations of space and each consen-
Eq. (10), the recommendation modification plan E 1m sus round involving the same processes, we now
is provided to dm. Then, each DM will be assigned a skip several consensus rounds and go straight to the
random number. If the random number is less than last consensus round. The DMs’ weights for Case A
PA ¼ 1h, we assume that dm will accept E 1m ; in different consensus rounds are shown in Figure
otherwise, we suppose that dm will reject E 1m ; mean- 4. GCD and PO in each consensus round are visual-
while, dm can express his/her evaluation according ized in Figure 5.
^ 1 : By combining E
to his/her will in E 1 with E ^ 1 , E1
m
can be obtained. 5.1.3. Last consensus round
As with the first consensus round, two categories of
5.1.2.2 Evolution and management of public opin- agents are considered. In Case A and Case B, four
ion. For Case A, CL>CL: As agents are highly con- and six consensus rounds are required, respectively,
fident, Management Measure I will be adopted. By to reach the two thresholds.
analyzing agents’ individual opinions, the agents
who posted false information are exposed. The trust 5.1.3.1 Recommendation-feedback process.
,g ¼
1
values toward them are reduced to t Applying Eq. (10), recommendation modification
ð1q1 Þt, g , and the deducted part of the trust val-
0
plans can be obtained. When faced with a recom-
ues q1 t
0
, g is equally divided among the agents mendation modification plan E, a DM can choose
whose individual opinions o0 0: Applying Eqs. (2) to accept or reject it. Their evaluations can then be
and (7), opinions of agents except DMs and opin- collected by E and E:
^ For Case A, by combining E 4
ions of DMs in the first consensus round are com- with E^ , E can be obtained. For Case B, combining
4 4
Figure 6. Number of consensus rounds in Cases A and B. (a) Case A, (b) Case B.
Figure 7. Average values of IGCD in simulation I. (a) IGCD, (b) N, and (c) h.
Figure 9. Average values of PO under different values of CL in simulation II. (a) PO (CL ¼ 0.1), (b) PO (CL ¼ 0.3),
(c) PO (CL ¼ 0.5).
Figure 10. Average values of PO under different values of q1 in simulation II. (a) PO (q1 ¼ 0.1), (b) PO (q1 ¼ 0.3), and (c)
PO (q1 ¼ 0.5).
seen in Figure 11, where Figures 9 and 10 can be manage public opinion effectively. The main idea of
obtained by taking cross-sections. simulation method III is described as follows.
In the CRP, if agents are unsatisfied with the
5.2.3. Simulation III result of the LGDM event, they may post some false
Similar to simulation II, simulation III was designed information to guide others against the LGDM
to examine whether Management Measure II can event. Through timely, effective communication
16 G.-R. YANG ET AL.
with DMs who often own numbers of followers, In simulation method III, we set different input
they are willing to raise their support degree to parameters q2 ¼ f0, 0:05, :::, 0:3g and CL ¼
ensure that the LGDM event can go smoothly. By f0:1, 0:2, :::, 0:5g: Other relevant parameters are
replacing the input and Step 12 in simulation same as Simulation II. Then, we run the method
method II, we can obtain simulation method III, 1000 times to produce average values of PO. The
which is described in Algorithm IV. results of this process are presented in
Figures 12–14.
Comparing the results shown in Figures 13a
and 13c, we can infer that when agents have low
confidence levels, Management Measure II is more
effective, and the effectiveness of this measure in
managing agents with high confidence levels is
quite limited. The results represented in Figure 14
show that public opinion PO can be improved
more quickly with a large value of q2. Figure 12
reveals the influence of CL and q2 on the effi-
ciency of Management Measure II, in which
Figures 13 and 14 can be obtained by taking
cross-sections.
Figure 12. Average values of PO in simulation III.
Figure 13. Average values of PO under different values of CL in simulation III. (a) PO (CL ¼ 0.1), (b) PO (CL ¼ 0.3), and
(c) PO (CL ¼ 0.5).
Figure 14. Average values of PO under different values of q2 in simulation III. (a) PO (q2 ¼ 0.1), (b) PO (q2 ¼ 0.2), and (c)
PO (q2 ¼ 0.3).
Figure 15. Average values of PO and GCD. (a) CL ¼ 0.1 and (b) CL ¼ 0.5.
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 17
In practice, with the development of the deci- Information Fusion, 59, 84–102. https://doi.org/10.1016/
sion-making event, agents may make new friends j.inffus.2020.01.006
with each other. So, it would be interesting to Dong, Y., Ding, Z., Martnez, L., & Herrera, F. (2017).
Managing consensus based on leadership in opinion
study the influence of updating of relationships dynamics. Information Sciences, 397, 187–205.
among agents on the evolution of their opinions. Dong, Y., Li, C., Xu, Y., & Gu, X. (2015). Consensus-
based group decision making under multi-granular
unbalanced 2-tuple linguistic preference relations.
Notes Group Decision and Negotiation, 24(2), 217–242.
1. The public opinion of a decision-making event refers https://doi.org/10.1007/s10726-014-9387-5
to the aggregation of opinions of people including Dong, Y., Zhang, H., & Herrera-Viedma, E. (2016).
DMs, stakeholders, and others who have expressed Integrating experts’ weights generated dynamically into
opinions towards the event. the consensus reaching process and its applications in
2. The consensus refers to the extent to which all DMs managing non-cooperative behaviours. Decision
reach an agreement on a decision-making event. Support Systems, 84, 1–15. https://doi.org/10.1016/j.dss.
3. The false information refers to biased opinions or 2016.01.002
opinions which present objection too forcefully. Du, Z. J., Yu, S. M., & Xu, X. H. (2020). Managing non-
cooperative behaviours in large-scale group decision-
making: Integration of independent and supervised
Acknowledgements consensus-reaching models. Information Sciences, 531,
119–138. https://doi.org/10.1016/j.ins.2020.03.100
We acknowledge the financial support of the National Durrett, R., Gleeson, J. P., Lloyd, A. L., Mucha, P. J., Shi,
Natural Science Foundation of China (Grant Nos. F., Sivakoff, D., Socolar, J. E. S., & Varghese, C. (2012).
72001025 and 71772136), the China Postdoctoral Science Graph fission in an evolving voter model. Proceedings of
Foundation (Grant No. 2020M680017), the introduction the National Academy of Sciences of Sciences, 109(10),
project of China Postdoctoral international exchange pro- 3682–3687. https://doi.org/10.1073/pnas.1200709109
gram (Grant No. YJ20200266), the Strategic Research Friedkin, N. E., & Johnsen, E. C. (1990). Social influence
Grants of the City University of Hong Kong (Grant Nos. and opinions. The Journal of Mathematical Sociology,
7005193 and 7005380), and the Beijing Institute of 15(3–4), 193–206. https://doi.org/10.1080/0022250X.
Technology Research Fund Program for Young Scholars. 1990.9990069
Gou, X., Xu, Z., Liao, H., & Herrera, F. (2020). Consensus
model handling minority opinions and noncooperative
Disclosure statement
behaviours in large-scale group decision-making under
No potential conflict of interest was reported by double hierarchy linguistic preference relations. IEEE
the authors. Transactions on Cybernetics, 51(1), 283–296.
Hegselmann, R., & Krause, U. (2002). Opinion dynamics
and bounded confidence models, analysis and simula-
ORCID tion. Journal of Artificial Societies and Social
Simulation, 5(3), 1–2.
Guo-Rui Yang http://orcid.org/0000-0001-8189-0946
Liang, D., and Yi, B., & Xu, Z. (2020). Opinion dynamics
Ru-Xi Ding http://orcid.org/0000-0001-5414-0728
based on infectious disease transmission model in the
Jingjun (David) Xu http://orcid.org/0000-0002-
non-connected context of Pythagorean fuzzy trust rela-
9875-7620
tionship. Journal of the Operational Research Society, 1–21.
Li, C. C., Dong, Y., & Herrera, F. (2019). A consensus
References model for large-scale linguistic group decision making
with a feedback recommendation based on clustered
Chao, X., Kou, G., Peng, Y., & Herrera-Viedma, E. personalized individual semantics and opposing consen-
(2021a). Large-scale group decision-making with non- sus groups. IEEE Transactions on Fuzzy Systems, 27(2),
cooperative behaviours and heterogeneous preferences: 221–233. https://doi.org/10.1109/TFUZZ.2018.2857720
An application in financial inclusion. European Journal Li, Y., Kou, G., Li, G., & Wang, H. (2021). Multi-attribute
of Operational Research, 288(1), 271–293. https://doi. group decision making with opinion dynamics based
org/10.1016/j.ejor.2020.05.047 on social trust network. Information Fusion, 75,
Chao, X., Kou, G., Peng, Y., Herrera-Viedma, E., & 102–115. https://doi.org/10.1016/j.inffus.2021.04.010
Herrera, F. (2021b). An efficient consensus reaching Li, K., Liang, H., Kou, G., & Dong, Y. (2020a). Opinion
framework for large-scale social network group deci- dynamics model based on the cognitive dissonance: An
sion making and its application in urban resettlement. agent-based simulation. Information Fusion, 56, 1–14.
Information Sciences, 575, 499–527. https://doi.org/10. https://doi.org/10.1016/j.inffus.2019.09.006
1016/j.ins.2021.06.047 Li, G., Kou, G., Li, Y., & Peng, Y. (2020b). A group deci-
Degroot, M. H. (1974). Reaching a consensus. Journal of sion making approach for supplier selection with
the American Statistical Association, 69(345), 118–121. multi-period fuzzy information and opinion interaction
https://doi.org/10.1080/01621459.1974.10480137 among decision makers. Journal of the Operational
Ding, R.-X., Palomares, I., Wang, X., Yang, G.-R., Liu, B., Research Society, 1–14.
Dong, Y., Herrera-Viedma, E., & Herrera, F. (2020). Lin, C., Kou, G., Peng, Y., & Alsaadi, F. (2020).
Large-Scale decision-making: Characterization, tax- Aggregation of the nearest consistency matrices with
onomy, challenges and future directions from an artifi- the acceptable consensus in AHP-GDM. Annals of
cial intelligence and applications perspective. Operations Research, 1–17.
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 19
Liu, X., Xu, Y., & Herrera, F. (2019c). Consensus model Urena, R., Chiclana, F., Fujita, H., & Herrera-Viedma, E.
for large-scale group decision making based on fuzzy (2015). Confidence-consistency driven group decision
preference relation with self-confidence: Detecting and making approach with incomplete reciprocal intuition-
managing overconfidence behaviours. Information istic preference relations. Knowledge-Based Systems, 89,
Fusion, 52, 245–256. https://doi.org/10.1016/j.inffus. 86–96. https://doi.org/10.1016/j.knosys.2015.06.020
2019.03.001 Wu, Z., & Xu, J. (2012). Consensus reaching models of
Liu, B., Zhou, Q., Ding, R., Ni, W., & Herrera, F. (2019a). linguistic preference relations based on distance func-
Defective alternatives detection-based multi-attribute tions. Soft Computing, 16(4), 577–589. https://doi.org/
intuitionistic fuzzy large-scale decision making model. 10.1007/s00500-011-0756-6
Knowledge-Based Systems, 186, 104962. https://doi.org/ Xu, X., Du, Z., Chen, X., & Cai, C. (2019). Confidence
10.1016/j.knosys.2019.104962 consensus-based model for large-scale group decision
Liu, B., Zhou, Q., Ding, R. X., Palomares, I., & Herrera, making: A novel approach to managing non-coopera-
F. (2019b). Large-scale group decision making model tive behaviours. Information Sciences, 477, 410–427.
based on social network analysis: Trust relationship- https://doi.org/10.1016/j.ins.2018.10.058
based conflict detection and elimination. European Xu, X., Zhang, Q., & Chen, X. (2020). Consensus-based
Journal of Operational Research, 275(2), 737–754. non-cooperative behaviours management in large-
https://doi.org/10.1016/j.ejor.2018.11.075 group emergency decision-making considering experts’
Lu, Y., Xu, Y., Herrera-Viedma, E., & Han, Y. (2021). trust relations and preference risks. Knowledge-Based
Consensus of large-scale group decision making in
Systems, 190, 105108. https://doi.org/10.1016/j.knosys.
social network: The minimum cost model based on
2019.105108
robust optimization. Information Sciences, 547,
Zha, Q., Kou, G., Zhang, H., Liang, H., Chen, X., Li, C.-
910–930. https://doi.org/10.1016/j.ins.2020.08.022
C., & Dong, Y. (2020). Opinion dynamics in finance
Palomares, I., Martinez, L., & Herrera, F. (2014). A con-
and business: A literature review and research opportu-
sensus model to detect and manage noncooperative
nities. Financial Innovation, 6(1), 1–22.
behaviours in large-scale group decision making. IEEE
Zhang, H., Li, C.-C., Liu, Y., & Dong, Y. (2019).
Transactions on Fuzzy Systems, 22(3), 516–530. https://
doi.org/10.1109/TFUZZ.2013.2262769 Modelling personalized individual semantics and con-
Ren, R., Tang, M., & Liao, H. (2020). Managing minority sensus in comparative linguistic expression preference
opinions in micro-grid planning by a social network relations with self-confidence: An optimization-based
analysis-based large scale group decision making approach. IEEE Transactions on Fuzzy Systems, 29(3),
method with hesitant fuzzy linguistic information. 627–640. https://doi.org/10.1109/TFUZZ.2019.2957259
Knowledge-Based Systems, 189, 105060. https://doi.org/ Zhang, H., Palomares, I., Dong, Y., & Wang, W. (2018).
10.1016/j.knosys.2019.105060 Managing non-cooperative behaviours in consensus-
Song, Y., Li, G., Ergu, D., & Liu, N. (2021). An optimisa- based multiple attribute group decision making: An
tion-based method to conduct consistency and consen- approach based on social network analysis. Knowledge-
sus in group decision making under probabilistic Based Systems, 162, 29–45. https://doi.org/10.1016/j.
uncertain linguistic preference relations. Journal of the knosys.2018.06.008
Operational Research Society, 1–15. Zhang, H., Zhao, S., Kou, G., Li, C.-C., Dong, Y., &
Tian, Z.-P., Nie, R.-X., & Wang, J.-Q. (2019). Social net- Herrera, F. (2020). An overview on feedback mecha-
work analysis-based consensus-supporting framework nisms with minimum adjustment or cost in consensus
for large-scale group decision-making with incomplete reaching in group decision making: Research para-
interval type-2 fuzzy information. Information Sciences, digms and challenges. Information Fusion, 60, 65–79.
502, 446–471. https://doi.org/10.1016/j.ins.2019.06.053 https://doi.org/10.1016/j.inffus.2020.03.001
Step 10.
Step 10: If POz 2 , go to Step 13; otherwise, go to Input: Zmax , X, A, D, R, T, C, h, 1, 2, and q1.
next step. Step 6: If z Zmax , output PO; otherwise, go to
Step 11: Calculate the average confidence level of agents CL. next step.
If CL CL, go to Step 12-A; otherwise, go to Step 12-B. Step 12: Management Measure I is adopted. For agent
Step 12-A: Management Measure I is adopted. For agent ag , if ozg <0, the trust values assigned to him/her t−zþ1,g
ag, if ozg <0, the trust values assigned to him/her t−zþ1 will be reduced to ð1−q1 Þt−z , g , and the deducted part of
, g will
be reduced to ð1−q1 Þt−z , g , and the deducted part of trust trust values q1 t−z , g will be equally divided among the
values q1 t−z , g will be equally divided among the agents agents whose individual opinions oz 0. Then, go to
whose individual opinions oz 0. Then, go to Step 13. Step 13.
Step 12-B: Management Measure II is adopted. For DM Output: PO.
dm who accept the recommendation modification plan,
/zm ¼ /zm þ q2 (if the adjusted /zm >1, then /zm ¼ 1).
Then, go to Step 13. Algorithm 4. Simulation method III.
Step 13: By using Eqs. (2) and (7), ozþ1 can be calculated.
Let z ¼ z þ 1,and go to Step 4. Input: Zmax , X, A, D, R, T, C, h, 1, 2, and q2.
Step 14: By applying Eq. (12), EV can be calculated. Step 12: Management Measure II is adopted. For DM dm
Step 15: Select and output the optimal alternative x . who accept the recommendation modification plan, /zm ¼
Step 16: The LGDM event failed. /zm þ q2 (if the adjusted /zm >1, then /zm ¼ 1). Then, go
Output: x . to Step 13.