On Leveraging Multi-Agent Models in Countermeasures for Deepfake Detection and Mitigation: A Comparative Analysis of Social Media Platform Strategies

Abstract

The proliferation of deepfake content on social media in recent years has posed significant threats to both individual and societal security. Consequently, devising effective countermeasures to curb the spread of deepfake information has become a critical challenge for social media platforms worldwide. This study aims to explore the propagation dynamics of deepfake information and evaluate the efficacy of various countermeasures by constructing a multi-agent model that integrates the SIR epidemiological model with the BA scale-free network theory. The research focuses on three distinct social media platforms in China—Douyin, Weibo, and Bilibili—as case studies. Through a series of simulation experiments, we compare the propagation patterns of deepfake content and analyze the performance differences of various countermeasures. The results indicate that, in terms of preventing the dissemination of deepfake information, the “preemptive defense” mechanism (exemplified by Douyin) proves to be the most effective in limiting the spread of deepfakes and ensuring timely counteractions. In contrast, the “post-verification” framework (exemplified by Weibo) is particularly effective in enhancing immunity against deepfake content. However, countermeasures based on self-media strategies that emphasize “emotion, viewpoints, and stances” (exemplified by Bilibili) demonstrate higher infection rates, weaker immunity, and longer response delays. The findings of this study offer valuable insights for developing more efficient and adaptive information governance strategies.