Preference similarity network structural equivalence clustering based consensus group decision making model

• Published in: Applied soft computing Vol. 67; pp. 706 - 720
• Main Authors: Kamis, Nor Hanimah, Chiclana, Francisco, Levesley, Jeremy
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.06.2018
• Subjects: Agglomerative hierarchical clustering; Consensus group decision making; IOWA-based aggregation operator; Opinion similarity; Social network analysis; Structural equivalence; Opinion similarity; Structural equivalence; Social network analysis; Consensus group decision making; IOWA-based aggregation operator; Agglomerative hierarchical clustering
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2017.11.022

[Display omitted] •Develop an undirected weighted preference network similarity measure.•Expresses strength of experts’ connections sharing most similar preferences.•An undirected weighted preference network structure is derived.•Utilized in the agglomerative hierarchical clustering with complete linkage function.•Group cluster consensus combine clusters’ internal and external cohesion indexes.•Feedback mechanism integrates both SNA and clustering methodologies. Social network analysis (SNA) methods have been developed to analyse social structures and patterns of network relationships, although they have been least explored and/or exploited purposely for decision-making processes. In this study, we bridge a gap between SNA and consensus-based decision making by defining undirected weighted preference network from the similarity of expert preferences using the concept of ‘structural equivalence’. Structurally equivalent experts are represented using the agglomerative hierarchical clustering algorithm with complete link function, thus intra-clusters’ experts are high in density and inter-clusters’ experts are rich in sparsity. We derive cluster consensus based on internal and external cohesions, while group consensus is obtained by identifying the highest level consensus at optimal level of clustering. Thus, the clustering based approach to consensus measure contributes to present homogeneity of experts preferences as a whole. In the event of insufficient group consensus state, we construct a feedback mechanism procedure based on clustering that consists of three main phases: (1) identification of experts that contribute less to consensus; (2) identification of a leader in the network; and (3) advice generation. We make use of the centrality concept in SNA as a way of determining the most important person in a network, who is presented as a leader to provide advices in the feedback process. It is proved that the implementation of the proposed feedback mechanism increases consensus and, because of the bounded condition of consensus measure, convergence to sufficient group agreement is guaranteed. The centrality concept is also applied in the construction of a new aggregation operator, namely as cent-IOWA operator, that is used to derive the collective preference relation from which the feasible alternative of consensus solution, based on the concept of dominance, is achieved according to a majority of the central experts in the network, which is represented in this paper by the linguistic quantifier ‘most of.’ For validation purposes, an existing literature study is used to perform a comparative analysis from which conclusions are drawn and explained.