[1G4-OS-26a] OS-26

[1G4-OS-26a-01] A System to Show Evidence for Reason of Dangerous Behaviors in Home Generated by Sentence Generation AI

〇Fumikatsu Anaguchi¹, Takeshi Morita^1,2 (1. Aoyama Gakuin University, 2. National Institute of Advanced Industrial Science and Technology)

Keywords:generative AI, Retrieval Augmented Generation, explainable AI, Large Language Model, Knowledge Graph

To use sentence generation AI safely and securely, it is necessary to show evidence citing the literatures. Therefore, in this work, we propose a system to show evidence for reason of dangerous behaviors in home generated by sentence generation AI for the dataset presented by Knowledge Graph Reasoning Challenge 2023.
First, we extract dangerous behaviors in home. Second, Sentence Generation AI generate reasons of it. Third, Retrieval Augmented Generation (RAG) retrieves a sentence similar to this reason from the literatures on dangerous behaviors in Home and shows the user as evidence.
We did a survey to evaluate whether the sentence generation AI can appropriately generate reasons for dangerous behaviors in the home, and whether the evidence showed by the proposed system for these reasons is appropriate.
As a result, average score of five-stage evaluation were 3.6 and 2.6. It is found that the proposed system can show general evidence for reasons.