Abstract:
Accidents among people over 65 years of age predominantly occur within residential settings, making the maintenance of a safe home environment a crucial social issue. To ...Show MoreMetadata
Abstract:
Accidents among people over 65 years of age predominantly occur within residential settings, making the maintenance of a safe home environment a crucial social issue. To address this issue, previous research has developed systems that construct Knowledge Graphs (KG) based on simulations of daily household activities, and studies have been conducted on detecting hazardous behaviors using such KG analysis. In this current study, we propose a system capable of presenting the reason and justification for the detected domestic hazardous behaviors. Our system will first generates the reason for the detected behavior using a Large Language Model (LLM). To ensure the accuracy, reliability and reproducibility of the LLM output, the system will provides reliable sources to support the output. We employed Retrieval-Augmented Generation (RAG) to search for sentences similar to the reason generated by the LLM within reliable, authoritative documents describing domestic accident cases and their causes and these will be presented as the evidence alongside the search engine results to the users. Consequently, a knowledge graph (KG) of domestic hazardous behavior is developed based on evidence ontology. Finally, to evaluate the ability of our proposed system in appropriately generating reasons for domestic hazardous behaviors and the adequacy of the justifications provided, the output was rated using LLMs and human volunteers. The rating results showed a significant correlation between LLMs and human evaluation, indicating that the proposed system can provide sufficient reasons and justifications for domestic hazardous behaviors at residential setting.
Date of Conference: 26-29 November 2024
Date Added to IEEE Xplore: 31 December 2024
ISBN Information: