A sentiment analysis-based two-stage consensus model of large-scale group with core-periphery structure

The development of big data and social media has driven large-scale group decision making (LSGDM) to merge with social networks and focus on individual behavioral factors. Following this trend, this paper develops a novel LSGDM consensus model that explores and manages the meso-scale structure among experts using free texts to express their opinions under social network settings. In the proposed approach, firstly the sentiment analysis is adopted to extract preferences over alternatives provided by experts and the preferences are further converted into distributed linguistic preference relation matrices. Then a coreperiphery detection method for the social network constructed based on the newly defined distance measure for linguistic distribution assessments is proposed. After that, expert weights are derived by an optimization model that maximizes the expert reliability based on consistency and node centrality. Moreover, considering reference dependence and bounded rationality features of members among the detected network, a prospect theory-based two-stage consensus model is developed to improve group consensus systematically and gradually. Finally, a case study regarding life science investments is provided to illustrate the usefulness of our proposal. The convergence of the proposed model is proven by theoretical and simulation analysis. Comparative analysis reveals the features and advantages of our model.