Reproduce “PRISM: A Personality-Driven Multi-Agent Framework for Social Media Simulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhixiang Lu, Xueyuan Deng, Yiran Liu, Yulong Li, Qiang Yan, Imran Razzak, Jionglong Su
PRISM, a personality-driven multi-agent framework, combines the Myers-Briggs Type Indicator with stochastic differential equations and partially observable Markov decision processes to simulate nuanced social media dynamics. The framework achieved a Polarity Error (MAE) of 0.14 and a behavioral consistency (ρ) of 0.782 across 1,000 controversial topics, accurately replicating emergent phenomena like rational suppression and affective resonance.
Key insights
- MBTI's categorical structure provides a robust taxonomy for computational simulation, allowing for the assignment of discrete and role-based interaction policies, which is computationally advantageous over continuous trait models.
- The interplay between rational suppression (Thinking types) and affective resonance (Feeling types) is critical for shaping opinion trajectories and maintaining sentiment fidelity in social dynamics, as evidenced by the significant impact of ablating the T/F dimension.
- By coupling continuous emotional dynamics (SDE) with discrete, personality-conditioned decision-making (PC-POMDP), the framework successfully resolves the "state amnesia problem" of traditional ABMs, enabling agents to maintain long-term behavioral coherence and dynamic emotional evolution.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.