The Dynamics of Emotional Chats with Bots: Experiment and Agent-Based Simulations

Bosiljka Tadic, Vladimir Gligorijevic, Marcin Skowron, Milovan SuvakovarXiv 2014 · OpenReview

Abstract

Quantitative research of emotions in psychology and machine-learning methods for extracting emotion components from text messages open an avenue for physical science to explore the nature of stochastic processes in which emotions play a role, e.g., in human dynamics online. Here, we investigate the occurrence of collective behavior of users that is induced by chats with emotional Bots. The Bots, designed in an experimental environment, are considered. Furthermore, using the agent-based modeling approach, the activity of these experimental Bots is simulated within a social network of interacting emotional agents. Quantitative analysis of time series carrying emotional messages by agents suggests temporal correlations and persistent fluctuations with clustering according to emotion similarity. {All data used in this study are fully anonymized.}

Introduction

The rise of social bots in online environments has introduced new variables into human social dynamics. While individual-level influence is well-documented, the macro-level emergence of collective behavior driven by these bots remains poorly understood. This study bridges statistical physics and social science to investigate if bots can act as emotional catalysts, altering the collective mood and structure of a social network.

Methods

The study utilized a hybrid approach combining empirical experiments and agent-based simulations:

  1. Experimental Design: 91 participants interacted with an "Affective Dialog System" (ADS) across three profiles: positive (empathic), negative (confrontational), and neutral (task-oriented).
  2. Linguistic Analysis: The ANEW (Affective Norms for English Words) dictionary was used to quantify the valence, arousal, and dominance of user messages.
  3. Agent-Based Modeling (ABM): Experimental data was integrated into a larger simulated network of interacting agents. The dynamics were governed by non-linear mathematical maps simulating emotional "charge" accumulation and decay.
  4. Statistical Analysis: Network modularity was used for community detection, and the Hurst exponent was calculated to assess the persistence of emotional fluctuations.

Results

The experimental bots successfully induced the intended emotional states in users. Linguistic analysis via the ANEW classifier showed significant differences in the valence and arousal of messages depending on the bot profile.

Mean ANEW Classifier ratings for valence, arousal, and dominance across the three bot conditions.

Temporal analysis of the interaction logs revealed that these emotional states were not transient but exhibited persistent fluctuations and spatial clustering.

Temporal valence dynamics (a) and user similarity heatmap (b) showing emotional clustering.

The calculated Hurst exponent ($H \approx 0.614$) confirms that the emotional dynamics possess long-term memory, meaning a "mood" in a conversation tends to sustain itself rather than reverting immediately to a neutral mean. In the simulated network, positive bots were found to trigger large-scale "emotional avalanches," while negative bots caused the highest degree of community instability, leading users to switch emotional clusters more frequently due to the combined pressure of the bot and their peers.

Discussion

This research demonstrates that bots can effectively anchor and shift the emotional state of both individuals and collective networks. The persistence of these emotional fluctuations suggests that social networks act as amplifiers for bot-induced stimuli. Future work should incorporate more sophisticated LLM-based bots and account for individual psychological traits to better predict and manage the spread of collective emotions online.

Conclusion

By combining experimental psychology with agent-based physics models, this study provides a framework for understanding how artificial agents can influence the self-organization of human emotions in digital environments.

Summary updated

Abstract

Quantitative research of emotions in psychology and machine-learning methods for extracting emotion components from text messages open an avenue for physical science to explore the nature of stochastic processes in which emotions play a role, e.g., in human dynamics online. Here, we investigate the occurrence of collective behavior of users that is induced by chats with emotional Bots. The Bots, designed in an experimental environment, are considered. Furthermore, using the agent-based modeling approach, the activity of these experimental Bots is simulated within a social network of interacting emotional agents. Quantitative analysis of time series carrying emotional messages by agents suggests temporal correlations and persistent fluctuations with clustering according to emotion similarity. {All data used in this study are fully anonymized.}
BibTeX
@misc{tadic2014dynamics,
  author = {Bosiljka Tadic and Vladimir Gligorijevic and Marcin Skowron and Milovan Suvakov},
  title = {{The Dynamics of Emotional Chats with Bots: Experiment and Agent-Based Simulations}},
  year = {2014},
  eprint = {1404.4191},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1404.4191},
}
Figure