Werewolf Among Us
收藏资源简介:
Werewolf Among Us是一个多模态数据集,旨在模拟社交推理游戏中的说服行为。该数据集由佐治亚理工学院、上海交通大学、Meta AI和斯坦福大学合作创建,包含199个多玩家社交推理游戏场景中的对话转录和视频记录。数据集不仅提供了游戏级别的推理游戏结果标注,还包含了26,647条说服策略级别的详细标注。创建过程中,研究者们利用了Ego4D社交数据集和YouTube视频,确保了数据的自然性和多样性。该数据集主要用于研究说服策略如何影响社交互动中的推理结果,以及如何通过对话上下文和视觉信号来预测说服策略。
Werewolf Among Us is a multimodal dataset developed to simulate persuasive behavior in social deduction games. It was collaboratively constructed by Georgia Institute of Technology, Shanghai Jiao Tong University, Meta AI, and Stanford University. The dataset includes dialogue transcripts and video recordings from 199 multiplayer social deduction game sessions. In addition to game-level annotations of deduction game outcomes, it also features detailed annotations for 26,647 persuasive strategy instances. During the dataset creation process, researchers utilized the Ego4D social dataset and YouTube videos to guarantee the naturalness and diversity of the data. This dataset is primarily used to investigate how persuasive strategies impact deduction outcomes in social interactions, as well as how to predict persuasive strategies using conversational context and visual signals.
- 1Werewolf Among Us: A Multimodal Dataset for Modeling Persuasion Behaviors in Social Deduction Games佐治亚理工学院, 上海交通大学, Meta AI, 斯坦福大学 · 2022年



