关键词: 多属性决策, 共识反馈模型, 情感分析, 可信度, 新冠疫情

Abstract: A group consensus decision-making method is proposed to adjust expert opinions based on public opinions in a social network environment. A sentiment analysis and TF-IDA technology is used to process public opinions on social media platforms and determine the number of attributes and attribute weights. Experts evaluate with linguistic evaluation values to form decision matrices, which are transformed into intuitionistic fuzzy decision matrices. Then the credibility degrees of experts are calculated from intuitionistic fuzzy decision matrices and the collective credibility degree matrix is calculated by using attribute weights. The consensus degree of expert is used to calculate weights of decision makers, consensus threshold and confidence threshold. An expert who has a low consensus degree should revise his/her evaluation values to decrease the deviation to other experts. If the expert agrees to revise his/her evaluation values, the other expert should be found who has the highest similarity degree with him/her and has reached consensus. Then mathematical programming models are set up to calculate the minimum adjustment cost. The experts who refuse to adjust should provide reasons for rejection and other experts give their degrees of recognition. If the other experts approve the rejection reasons, they will adjust their opinions. Otherwise, the weight of the expert will be reduced through a reduction coefficient. This process is repeated until a consensus is reached. Finally, the comprehensive evaluation values of the alternatives are calculated using the intuitionistic fuzzy weighted average operator and ranked accordingly.
The new proposed method has been illustrated by the management problem of the COVID-19. First public comments are collected from Weibo in the pandemic situation including Xi’an, Shanghai, Chengdu, Beijing for aspects of resident lives, epidemic prevention measures, government management, community management. Sentiment analysis has been used to deal with public opinions to derive intuitionistic fuzzy public preference matrix. Then the new consensus-feedback decision-making model has been used to rank the four cities. The results demonstrate that the proposed method effectively enhances the credibility of decision-makers’ opinions. Furthermore, conflicting opinions acknowledged by decision experts are preserved to make the proposed method more objective and accurate.
Compared with existing research on consensus-feedback problems, public opinions are considered to get attributes of evaluation problems and reference information for experts in interactive process, which can improve public participation in the decision-making process and decision results are easier to be accepted by the public. In consensus process, the experts are allowed to give their refusal reasons if they refuse to modify their evaluation values and other experts give their opinions to decide whether to be accepted or not, which can respect the opinion of experts and improve the quality the decision results.
In the context of group consensus decision making in social network, there are still many issues worth further studying such as impact of dynamic public opinions, the computation complexity of large-scale group decision making problems, influence of different types of experts in decision making process, etc. Future research will focus on how to extend the proposed method to deal with dynamic decision-making problems.

Key words: multi-attribute decision making, consensus feedback model, sentiment analysis, credibility, COVID-19