Subjective-Graph LLM Agents for Simulating Uncertainty in Classroom Social Perception
Jinming Yang, Xinyu Jiang, Xinshan Jiao, Xinping Zhang
Why It Matters
What makes this one worth your time
Understanding how subjective perceptions affect social dynamics in educational settings can inform interventions to improve academic and social outcomes.
The study models classroom social perception using subjective-graph LLM agents to explore persistent perception distortions.
Summary
The paper introduces a multi-agent framework using subjective-graph LLM agents to simulate social perception in classrooms, focusing on how individual perceptions can lead to persistent distortions in perceived academic standing despite objective performance signals. The framework is evaluated on middle-school classrooms, showing increased collective ranking error over time and demonstrating the benefits of individualized visibility and trust gating.
Key contributions
- Introduction of a data-constrained multi-agent framework using subjective-graph LLM agents.
- Evaluation of the framework on real classroom data, showing increased ranking error over time.
- Comparison with DeGroot configurations, demonstrating lower final ranking error and maintained opinion diversity.
Notable insights
- The use of subjective graphs and credibility-weighted communication can lead to persistent distortions in social perception.
- Individualized visibility and LLM-based trust gating contribute to more stable long-term behavior in social perception simulations.
Possible limitations
- Not stated in the abstract
Abstract
arXiv:2603.20750v2 Announce Type: replace Abstract: Social actors do not observe a common social world: each individual forms judgments from a partial and potentially distorted view of the surrounding network. We study whether graph-local evidence and credibility-weighted communication can generate persistent distortions in perceived academic standing, even when agents repeatedly receive objective performance signals. We introduce a data-constrained multi-agent framework in which LLM agents operate through individualized subjective graphs that determine peer visibility, evidence access, and interaction opportunities. Agents exchange uncertainty-annotated assessments, evaluate message credibility, and maintain explicit Gaussian belief states updated through Bayesian fusion. We evaluate the framework on 12 middle-school classrooms comprising 482 students, using questionnaire-derived social information and six consecutive examinations. On the Social-Observed subset (n=419), collective ranking error increases from 0.066 \pm 0.008 to 0.124 \pm 0.009 across six epochs despite repeated exam-based anchoring. Ablations associate individualized visibility and LLM-based trust gating with more stable long-horizon behavior, while constrained retrieval primarily safeguards against global-information leakage. Compared with evaluated DeGroot configurations, the proposed framework achieves lower final ranking error; those DeGroot configurations exhibit near-zero terminal opinion diversity. These findings establish subjective-graph LLM agents as a mechanism-oriented framework for data-constrained simulated social perception. Code is available at https://anonymous.4open.science/r/Rashomonomon-0126.