NegotiationToM

NegotiationToM是一个由香港科技大学开发的新型基准数据集，专门设计用于测试机器在真实世界谈判环境中的心智理论（ToM）能力。该数据集基于Belief-Desire-Intention（BDI）代理模型理论，覆盖了多维心理状态，如愿望、信念和意图。数据集包含395个对话，共计2,380轮谈判和4,618条发言。每个发言都附带了关于愿望、信念和意图的七个问题和标注标签，旨在评估大型语言模型（LLMs）在理解对话参与者心理状态方面的能力，特别是在信息逐渐增多的对话轮次中的表现。NegotiationToM的应用领域主要集中在评估和提升机器在复杂社会互动中的理解和决策能力，特别是在需要深入理解他人心理状态的谈判场景中。

NegotiationToM is a novel benchmark dataset developed by The Hong Kong University of Science and Technology, specifically designed to test the theory of mind (ToM) capabilities of machines in real-world negotiation scenarios. Grounded in the Belief-Desire-Intention (BDI) agent model theory, this dataset covers multidimensional mental states including desires, beliefs and intentions. It contains 395 dialogues, totaling 2,380 negotiation turns and 4,618 utterances. Each utterance is accompanied by seven questions and annotated labels related to desires, beliefs and intentions, which aims to evaluate the ability of large language models (LLMs) to understand the mental states of dialogue participants, especially their performance across dialogue turns as information gradually accumulates. The application domains of NegotiationToM mainly focus on evaluating and enhancing the understanding and decision-making capabilities of machines in complex social interactions, particularly in negotiation scenarios that require in-depth understanding of others' mental states.