A trust induced recommendation mechanism for reaching consensus in group decision making

• Published in: Knowledge-based systems 2017-03, Vol.119, p.221-231
• Main Authors: Liu, Yujia, Liang, Changyong, Chiclana, Francisco, Wu, Jian
• Format: Article
• Language: English
• Subjects: Decision making; Group consensus; Group decision making; Harmony degree; Recommendation mechanism; Recommender systems; Trust; Trust degree; Trustworthiness; Uncertainty
• ISSN: 0950-7051; 1872-7409
• DOI: 10.1016/j.knosys.2016.12.014

•The visual trust relationship is constructed.•A trust induced recommendation mechanism is investigated.•It arrives at the threshold value with the high harmony degree simultaneously.•An interval-valued trust decision making space is developed to model uncertainty. This article addresses the inconsistency problem in group decision making caused by disparate opinions of multiple experts. To do so, a trust induced recommendation mechanism is investigated to generate personalised advices for the inconsistent experts to reach higher consensus level. The concept of trust degree (TD) is defined to identify the trusted opinion from group experts, and then the visual trust relationship is built to help experts ‘see’ their own trust preferences within the group. Consequently, trust based personalised advices are generated for the inconsistent experts to revisit their opinions. To model the uncertainty of experts, an interval-valued trust decision making space is defined. It includes the novel concepts of interval-valued trust functions, interval-valued trust score (IVTS) and interval-valued knowledge degree (IVKD). The concepts of consensus degree (CD) between an expert and the rest of experts in the group as well as the harmony degree (HD) between the original opinion and the revised opinion are developed for interval-valued trust functions. Combining HD and CD, a more reasonable policy for group consensus is proposed as it should arrive at the threshold value with the maximum value of harmony and consensus degrees simultaneously. Furthermore, because the trust induced recommendation mechanism focuses on changing inconsistent opinions using only opinions from the trusted experts and not from the distrusted ones, the HD based changes cost to reach the threshold value of consensus is lower than previous mechanisms based on the average of the opinion of all experts. Finally, once consensus has been achieved, a ranking order relation for interval-valued trust functions is constructed to select the most appropriate alternative.