Game Data for One Night Ultimate Werewolf

该数据集包含了在《一夜终极狼人》环境中，使用LLM智能体进行竞争的游戏数据，重点在于整合多模态线索和心智理论。具体来说，该数据集不仅包含了面部表情和语音音调等多模态数据，还包含了智能体对智能体模拟以及人类玩家互动的游戏结果。数据集规模为5300局游戏（其中5000局用于训练，300局用于验证），其任务是对社交推理游戏进行分析以及模型训练。This dataset comprises game data from competitive matches involving LLM-based AI Agents within the setting of *One Night Ultimate Werewolf*, with a core focus on integrating multi-modal cues and theory of mind. Specifically, the dataset includes not only multi-modal data such as facial expressions and voice pitch, but also game outcomes derived from both agent-to-agent simulations and interactions between AI agents and human players. The dataset totals 5,300 game sessions, of which 5,000 are allocated for model training and 300 for validation, and it is designed for social deduction game analysis and model training.

搜集汇总

背景与挑战

背景概述

该数据集聚焦于《一夜终极狼人》游戏环境，收集了使用LLM智能体进行竞争的游戏数据，重点整合多模态线索（如面部表情和语音音调）和心智理论，用于社交推理分析。数据集包含5300局游戏，其中5000局用于训练、300局用于验证，支持智能体与人类玩家互动的游戏结果研究，适用于模型训练和社交行为模拟。