A method of predicting and managing public opinion on social media: An agent-based simulation

Abstract

In current opinion dynamics models for predicting public opinion, the spread of events within social media has been inadequately considered, resulting in suboptimal prediction performance and inefficient strategies for public opinion management. This deficiency is particularly consequential for governments and enterprises, as adverse public opinions associated with them can inflict significant harm. This study develops a link prediction-based opinion dynamics (LPOD) model to address this gap in predicting and managing public opinion. The proposed model integrates insights from epidemiology, specifically the susceptible-infected-recovered model, to characterize the spread of events. The LPOD model enhances the updating process of relationships and opinions by redesigning the link and opinion prediction methods. Subsequently, a link-recommendation-based management approach is formulated to manage public opinion effectively. Experimental results reveal that, compared to existing models, the proposed model elevates opinion prediction accuracy from 0.81 to 0.92 and link prediction accuracy from 0.60 to 0.70. In terms of public opinion management efficacy, when compared to conventional methods such as managing opinion leaders and introducing particular nodes in social networks, the developed approach demonstrates a 20% and 18% increase in success rates, respectively. Furthermore, validation through simulations and real-world scenarios robustly confirms the model's versatile applicability and effectiveness.