At the 2025 Electrical Society Electronics/Information/Systems Division Conference held at the Kanazawa Institute of Technology Ogigaoka Campus from 2025/8/27 to 8/30, Yasumi Yuai, a 4th year student in the Department of Applied Information Engineering belonging to Takahashi Laboratory, received the Excellent Poster Award.
The title of the presentation is “Proposal of a RAG Response System Considering the Speaker's Level of Understanding” (Yasumi Yuai, Sato Shunichi, Fukuda Takumi, Takahashi Hijiri). In this study, we constructed a system that generates responses while referring to external knowledge based on search extension generation (RAG), a technology for dealing with halcination problems in large-scale language models (LLM) represented by ChatGPT. Furthermore, it is a study that suppresses redundancy in conversation by estimating the user's level of understanding based on the level of difficulty of vocabulary and sentence structure, and introducing a mechanism that dynamically adjusts the simplicity and expertise of explanations according to that level. As a result of the evaluation, it was confirmed that the comprehension estimate was reasonable, and it was shown that an appropriate response expression can be generated for each user. Congratulations on winning the award.Proposal of a RAG response system that takes into account the speaker's level of understanding