主题：Rethinking the Micro-Foundation of Opinion Dynamics: Rich Consequences of an Inconspicuous Change

主讲人：梅文俊（北京大学工学院力学与工程科学系助理教授）

曾是瑞士联邦理工学院自动控制实验室博士后研究员，先后获得北京大学理论与应用力学学士、加州大学圣塔芭芭拉分校机械工程硕士和博士学位。他目前担任Journal of Mathematical Sociology的编委。他的主要研究领域包括社会和经济网络、网络博弈、群体决策和演化博弈。

时间：2021年10月20日11:00-12:30

地点：北京大学理科五号楼521

工作语言：中文

主办单位：北京大学社会研究中心 北京大学社会科学学部

Public opinion formation has long been affected by various harmful phenomena like opinion radicalization, echo chambers, and information manipulation. To identify the main mechanisms underlying complex opinion formation processes, researchers have long been exploring simple mechanistic mathematical models. Despite important progress in recent decades, many widely-studied models do not fully explain some important real-world phenomena including, e.g., opinion formation over ordered discrete options, continuous multimodal public opinion distributions, and radicalization of socially marginalized individuals. In this paper, we propose a parsimonious yet predictive model that sheds new light on these important phenomena. Remarkably, this new model is not constructed by introducing additional assumptions or parameters to previous models, but rather by changing the very core mechanism of opinion evolution. Specifically, we modify the widely adopted weighted-averaging mechanism and, inspired by cognitive dissonance theory, propose a novel micro-foundation for opinion dynamics, namely the weighted-median mechanism. This inconspicuous microscopic change leads to rich macroscopic consequences. The new mechanism broadens the applicability of opinion dynamics models to multiple-choice issues with ordered discrete options, as appearing for example in political elections. As shown by quantitative predictability tests on a large-scale online experimental dataset, predictions of opinion shifts by the median-based mechanisms enjoy significantly lower error rates than predictions by the corresponding averaging-based mechanisms. Moreover, comparative numerical studies indicate that, the weighted-median mechanism, despite its simplicity in form, is consistent with empirical observations in many important sociological phenomena. In addition, theoretical analysis shows that the weighted-median model exhibits rich dynamical behavior depending upon subtle yet robust network structures. All these results establish that the weighted-median mechanism is a viable new micro-foundation of opinion dynamics.