Impact of mindset types and social community compositions on opinion dynamics: A large language model-based multi-agent simulation study

Abstract

This study examines the impact of individual mindsets and social community compositions (SCC) on opinion dynamics through a large language model-based multi-agent simulation. We categorized mindsets into five types: very negative, more negative, neutral, more positive, and very positive, and simulated four SCC: uniformly distributed, normally distributed, negatively power-law distributed, and positively power-law distributed. Our investigation focused on opinion shifts regarding increased AI use in classrooms. Findings reveal that, compared to individuals with neutral and positive mindsets, those with negative mindsets experienced a greater degree of perspective change when influenced by others. They also exhibited a stronger tendency toward conformity, whereas moderately negative and positive nodes showed more opinion stability. Moreover, positive viewpoints were more effective in causing this change than neutral ones. The dominant mindset type within a community significantly shapes the public opinion environment. Additionally, individuals’ emotional tendencies towards a topic showed a moderate positive correlation with the number of positive arguments and a moderate negative correlation with the number of negative arguments. The use of large language models for simulating complex opinion formation processes in social networks represents a novel contribution to the field. These insights have important implications for understanding and managing public opinion in digital spaces, providing a foundation for future studies on opinion evolution in online communities.