Item type |
Symposium(1) |
公開日 |
2021-11-06 |
タイトル |
|
|
タイトル |
Prediction of Werewolf Players by SentimentAnalysis of Game Dialogue in Japanese |
タイトル |
|
|
言語 |
en |
|
タイトル |
Prediction of Werewolf Players by SentimentAnalysis of Game Dialogue in Japanese |
言語 |
|
|
言語 |
eng |
キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Imperfect Information Games |
キーワード |
|
|
主題Scheme |
Other |
|
主題 |
The Werewolf Game |
キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Sentiment Analysis |
キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Gated Recurrent Unit |
資源タイプ |
|
|
資源タイプ識別子 |
http://purl.org/coar/resource_type/c_5794 |
|
資源タイプ |
conference paper |
著者所属 |
|
|
|
Graduate School of Arts and Sciences, the University of Tokyo |
著者所属 |
|
|
|
Graduate School of Interdisciplinary Information Studies, the University of Tokyo |
著者所属(英) |
|
|
|
en |
|
|
Graduate School of Arts and Sciences, the University of Tokyo |
著者所属(英) |
|
|
|
en |
|
|
Graduate School of Interdisciplinary Information Studies, the University of Tokyo |
著者名 |
Yingxue, Sun
Tomoyuki, Kaneko
|
著者名(英) |
Yingxue, Sun
Tomoyuki, Kaneko
|
論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
The werewolf game is a communication game about trust and deception that is recently trending worldwide, and researchers have studied how to train AI agents to play this game. Compared to AI agents in other games, AI agents in the werewolf game need to speculate the intention of other players’ through their conversations in natural language in addition to other discrete actions. There are different viewpoints for agents in the werewolf game because agents can be assigned to different roles, leading to different information sets. This paper starts from a villager player’s viewpoint, and tries to analyze the only public information for them —— dialog using sentiment points, coming out information, and voting points. We implemented our models with Gated Recurrent Unit (GRU), and three tasks are evaluated via these data sets: werewolf prediction, prediction using CO information, and voting prediction. All of the tasks get accuracy better than the random model. |
論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
The werewolf game is a communication game about trust and deception that is recently trending worldwide, and researchers have studied how to train AI agents to play this game. Compared to AI agents in other games, AI agents in the werewolf game need to speculate the intention of other players’ through their conversations in natural language in addition to other discrete actions. There are different viewpoints for agents in the werewolf game because agents can be assigned to different roles, leading to different information sets. This paper starts from a villager player’s viewpoint, and tries to analyze the only public information for them —— dialog using sentiment points, coming out information, and voting points. We implemented our models with Gated Recurrent Unit (GRU), and three tasks are evaluated via these data sets: werewolf prediction, prediction using CO information, and voting prediction. All of the tasks get accuracy better than the random model. |
書誌情報 |
ゲームプログラミングワークショップ2021論文集
巻 2021,
p. 186-191,
発行日 2021-11-06
|
出版者 |
|
|
言語 |
ja |
|
出版者 |
情報処理学会 |