Reproduce “Think Thrice Before You Speak: Dual knowledge-enhanced Theory-of-Mind Reasoning for Persuasive Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Minghui Ma, Bin Guo, Runze Yang, Mengqi Chen, Yan Liu, Jingqi Liu, Yahan Pei, Xuehao Ma and 2 more
Researchers at Northwestern Polytechnical University developed "Think Thrice Before You Speak" (TTBYS), a framework to improve large language models' Theory-of-Mind (ToM) reasoning for persuasive agents by leveraging dual knowledge. This approach, which includes a new ToM-based Broad Persuasive Dialogues (ToM-BPD) dataset, enabled an open-source LLM (Qwen-3-8B) to surpass GPT-5 in accuracy for desire (72.82%), belief (54.64%), and strategy (39.78%) prediction.
Key insights
- Explicitly modeling the interdependent nature of mental states within a BDI framework, combined with a backward inference process, significantly improves the coherence and interpretability of ToM reasoning in LLM agents.
- Integrating diverse prior ToM experiences (both structured knowledge units and probabilistic distributions) effectively enhances LLMs' ability to infer complex mental states and select appropriate persuasive strategies.
- Smaller, open-source LLMs can achieve or surpass the performance of larger, closed-source models in specific, complex reasoning tasks when equipped with a structured, knowledge-enhanced reasoning framework.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.