Enhancing Dialogue Generation in Werewolf Game Through Situation Analysis and Persuasion Strategies

TL;DR

This paper introduces a LLM-based Werewolf Game AI, where each role is supported by situation analysis to aid response generation and various persuasion strategies are employed to effectively persuade other players to align with its actions.

Abstract

Recent advancements in natural language processing, particularly with large language models (LLMs) like GPT-4, have significantly enhanced dialogue systems, enabling them to generate more natural and fluent conversations. Despite these improvements, challenges persist, such as managing continuous dialogues, memory retention, and minimizing hallucinations. The AIWolfDial2024 addresses these challenges by employing the Werewolf Game, an incomplete information game, to test the capabilities of LLMs in complex interactive environments. This paper introduces a LLM-based Werewolf Game AI, where each role is supported by situation analysis to aid response generation. Additionally, for the werewolf role, various persuasion strategies, including logical appeal, credibility appeal, and emotional appeal, are employed to effectively persuade other players to align with its actions.

Figures (7)

• Figure 1: Proposed system architecture for the sUper_IL team's Werewolf Game AI. Before generating responses, all roles first utilize an LLM for situation analysis. The werewolf role uses logical appeal, credibility appeal, and emotional appeal to persuade other players' voting behavior.
• Figure 2: The prompt used for situation analysis.
• Figure 3: An example of generated situation analysis.
• Figure 4: The prompt used for generating responses for the seer role. The [CONDITION_ANALYSIS] section is generated by the LLM in the previous phase.
• Figure 5: The prompt used for persuasive response generation. Logical Appeal is used to urge other players to vote for Agent[03]. The section in blue is generated by the LLM in the previous phase.