He Said, She Said, They Simulated: Simulating Situated Gossip with LLM Agents

ACL ARR 2025 July Submission714 Authors

Abstract:

The pervasive nature of gossip in socialization underscores the need to understand its underlying mechanisms.
However, gossip behavior has been investigated primarily through vignette-based surveys, which provide limited insights into the dynamic and adaptive gossip behaviors within real-world contexts.
To address these limitations, we introduce an LLM agent-based simulation framework to realistically model social interactions within simulated workplace environments.
Our framework examines three key dimensions of gossip dynamics—initiation, reaction, and perception—grounded in established findings from prior research on gossip.
Empirical evaluations demonstrate that our simulation can reproduce established patterns in human gossip behavior, while also providing additional insights into the underlying reasoning and the temporally unfolding, context-sensitive behaviors.
Additionally, the framework's flexibility enables further analysis of multiple psychosocial and contextual factors that influence the dynamics of gossip.
This work aims to bridge LLM-based computational modeling with social science to advance the study of complex social interactions.

Paper Type: Long

Research Area: Computational Social Science and Cultural Analytics

Research Area Keywords: LLM Agents, Gossip simulation, Computational Social Science

Contribution Types: NLP engineering experiment

Languages Studied: English

Submission Number: 714