KIRINO : An Interactive Chatbot System for User Persona

Ganghun Kim1†, Hyunjae Kim2†, Geon Choi3†, Kyuhwan Shim4†, Myoung-Wan Koo5*
Sogang University, Computer Science and Engineering
Republic of Korea

†These authors contributed equally to this work.

*Corresponding author.

Paper Code

Project Architecture

RAG Architecture

Abstract

This research develops the KIRINO architecture, which utilizes large language models (LLMs) in conversational systems to learn the user's persona and generate personalized responses that reflect the user's tone and linguistic characteristics, combined with a retrieval-augmented generation (RAG) architecture. Unlike existing chatbots that simply answer user queries, the KIRINO architecture provides natural and consistent dialogues by reflecting the user's persona and manner of speech, thereby enhancing the quality of conversation and user experience. To address the issues of inconsistent responses, context deviation, and occasional dullness in existing chatbots, this system focuses on developing a persona-based conversational system. The proposed algorithm was evaluated using G-Eval and Human Eval, showing improvements of 26% and 38%, respectively, demonstrating significant enhancements in the quality and naturalness of conversations and contributing to the improvement of conversational interfaces.