New AI Framework Improves Theory of Mind Reasoning in Persuasive Dialogue
A new research paper introduces the ToM-PD task and TTBYS framework to enhance LLMs' Theory of Mind reasoning in persuasive dialogues. The work addresses LLMs' failure to capture dependencies among mental states by grounding reasoning in the Belief-Desire-Intention (BDI) framework. The authors constructed a large-scale annotated dataset called ToM-BPD, which captures fine-grained mental states and persuasive strategies. The proposed TTBYS framework leverages both explicit and implicit knowledge to improve stepwise reasoning. The paper is available on arXiv under ID 2605.22602.
Key facts
- The paper introduces the ToM-PD task for Theory of Mind reasoning in persuasive dialogue.
- The task is grounded in the Belief-Desire-Intention (BDI) framework.
- A large-scale annotated dataset called ToM-BPD was constructed.
- The TTBYS framework uses knowledge-enhanced stepwise reasoning.
- Existing LLMs fail due to reliance on simple prompting and insufficient ToM knowledge.
- The dataset captures fine-grained mental states and persuasive strategies.
- The paper is published on arXiv with ID 2605.22602.
- The research aims to improve persuasive agent capabilities.
Entities
Institutions
- arXiv