Unleashing Emotional Intelligence: How Role-Playing Agents Can Benefit from Advanced Emotional Retrieval

Emotional RAG: Enhancing Role-Playing Agents through Emotional Retrieval

cs.AI31 Oct 2024

Le Huang, Hengzhi Lan, Zijun Sun, Chuan Shi, Ting Bai

Beijing University of Posts and Telecommunications; Yunic.AI

As large language models (LLMs) exhibit high human capabilities, more and more attention has been devoted to the field of role-playing research, where the responses generated by LLMs are expected to mimic human replies. This has facilitated the exploration of role-playing agents in a variety of applications, such as chatbots capable of engaging in natural conversations with users and virtual assistants capable of providing personalized support and guidance. A key factor in role-playing tasks is the effective use of character memory, which stores the character’s profile, experiences, and historical dialogue. Retrieval-augmented generation (RAG) techniques are used to access relevant memories to enhance the role-playing agent’s response generation. Most existing research retrieves relevant information based on semantic similarity in memory to maintain the personality characteristics of characters, while few attempts to incorporate emotional factors into retrieval-enhanced generation (RAG) of LLMs. Inspired by the emotion-dependent memory theory, which suggests that people will have better recall results if they can re-experience the original emotion when they were learning, we propose a novel emotion-aware memory retrieval framework called emotion RAG, which The framework considers affective states in role-playing agents to recall relevant memories. Specifically, we design two retrieval strategies, namely, combinatorial strategy and sequential strategy, to combine memory semantics and affective states during retrieval. Extensive experiments on three representative role-playing datasets show that our emotional RAG framework outperforms methods that do not consider emotion factors in preserving role-playing agent personality. This provides evidence to further strengthen the emotion-dependent memory theory in psychology.