Selma: Simulating Personality-Driven Discussion Among Agents
Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment · 08.11.2026 · 22 (1) · 181-190
DOI: 10.1609/aiide.v22i1.43576 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, online erschienen am 08.11.2026. Verfasst von Octavio Villalobos und 3 weiteren Autor:innen.
Abstract (Originalsprache)
Existing multi-agent opinion simulations typically treat conversations as uniform exchanges: agents influence one another equally, discussions run for fixed turn limits, and nothing captures the reward or productivity of the exchange. We present Selma, a rule-based simulation of opinion dynamics in group conversation that grounds conversational behavior with agent personality. Each agent carries a set of attributes and a knowledge base that can be cited during interactions. Speaker selection, interaction type, and influence are all mediated by these attributes resulting in behavior such as charismatic agents speaking more and carrying greater persuasion, while uncertain agents are more easily moved. Conversations end when agents grow satisfied or frustrated with it, or when no progress in discussing a topic is made anymore. We validate our system's handling of opinion change, interaction patterns, conversation length and topics, and speaker clustering across a wide range of experiments. We show that satisfaction exits reliably reflect conversation tone, uncertain agents shift opinion substantially more than certain ones, and charisma concentrates speaking time, while observer participation contributes meaningfully to opinion change.
Zitieren
APA
Villalobos, O., Ernest, N. J., Pelyushenko, A., & Eger, M. (2026). Selma: Simulating Personality-Driven Discussion Among Agents. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, 22(1), 181-190. https://doi.org/10.1609/aiide.v22i1.43576
Vancouver
Villalobos O, Ernest NJ, Pelyushenko A, Eger M. Selma: Simulating Personality-Driven Discussion Among Agents. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment. 2026;22(1):181-190. doi:10.1609/aiide.v22i1.43576
BibTeX
@article{Villalobos2026,
author = {Villalobos, Octavio and Ernest, Nihal Jonathan and Pelyushenko, Alexei and Eger, Markus},
title = {Selma: Simulating Personality-Driven Discussion Among Agents},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment},
year = {2026},
volume = {22},
number = {1},
pages = {181--190},
doi = {10.1609/aiide.v22i1.43576}
}
EUVIMED