Social Intelligence Enhancement Methods for LLM Agents Based on Retrieval-Augmented Generation
Zhicheng Zhang, Wu Yang, Xiaolong Zheng · 2024
The rapid advancement of Large Language Models (LLMs) has enabled LLM agents to accomplish more complex tasks through interactions. However, in complex social interaction scenarios, LLM agents may still lack a certain degree of social intelligence. This paper proposes the automatic generation of an external repository of social skills by powerful LLMs like GPT-4 to guide agents in social interactions. Based on the Retrieval Augmented Generation (RAG) approach, the proposed method utilizing relevance-based extraction and random sampling to choose appropriate strategies from the social skills repository to enhance the content generation of LLMs. The results demonstrate that this approach can effectively improves the social intelligence of agents across a variety of social interaction scenarios.