Probabilistic Modeling of Intentions in Socially Intelligent LLM Agents

2025-10-21

AI Summary

Large language model (LLM) agents struggle to model interlocutors’ implicit intentions in multi-turn social dialogues due to inherent uncertainty and lack of explicit intent signals. Method: We propose a probabilistic intention modeling framework that dynamically estimates and maintains a belief distribution over the partner’s latent intentions via Bayesian belief updating—without requiring direct observation of ground-truth intentions. It combines context-informed prior initialization with turn-wise likelihood estimation to enable robust, uncertainty-aware intention inference. Contribution/Results: This work is the first to explicitly model belief distributions for social dialogue intention reasoning. Evaluated on the SOTOPIA benchmark, our framework achieves +9.0% and +4.1% improvements over the Qwen2.5-7B baseline on SOTOPIA-All and SOTOPIA-Hard, respectively—surpassing even an Oracle agent with access to true intent labels. The approach significantly enhances LLM agents’ contextual understanding and strategic adaptability in social interactions.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Multiagent Systems: Modeling other Agents
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Agentic search

Abstract

We present a probabilistic intent modeling framework for large language model (LLM) agents in multi-turn social dialogue. The framework maintains a belief distribution over a partner's latent intentions, initialized from contextual priors and dynamically updated through likelihood estimation after each utterance. The evolving distribution provides additional contextual grounding for the policy, enabling adaptive dialogue strategies under uncertainty. Preliminary experiments in the SOTOPIA environment show consistent improvements: the proposed framework increases the Overall score by 9.0% on SOTOPIA-All and 4.1% on SOTOPIA-Hard compared with the Qwen2.5-7B baseline, and slightly surpasses an oracle agent that directly observes partner intentions. These early results suggest that probabilistic intent modeling can contribute to the development of socially intelligent LLM agents.

Problem

• Modeling latent intentions in multi-turn social dialogue
• Updating belief distributions through probabilistic likelihood estimation
• Enabling adaptive dialogue strategies under uncertainty

Innovation

• Probabilistic framework models partner intentions in dialogue
• Belief distribution updated dynamically through likelihood estimation
• Adaptive dialogue strategies under uncertainty via contextual grounding