ChatASD: A Dialogue Framework for LLMs Enhanced by Autism Knowledge Graph Retrieval

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by developmental delays, communication difficulties, repetitive behaviors, and restricted interests. Large Language Models (LLMs) have demonstrated exceptional capabilities in various natural language tasks, particularly in providing personalized question-and-answer(Q&A) services, making them well-suited for constructing dialogue engines for autism Q&A systems. However, general LLMs often lack integrated autism knowledge during training, limiting their professional competency in autism consultation. Additionally, the automatic evaluation of scientific accuracy in autism medical knowledge Q&A remains underexplored. To address this gap, we propose ChatASD, an autism knowledge Q&A framework based on Graph Retrieval-Augmented Generation (GraphRAG) technology. This framework leverages LLMs and retrieves relevant information from medical literature to generate an autism knowledge graph, employing a combination of global and community queries to produce reliable responses. Compared to traditional methods, ChatASD effectively addresses the sparse distribution of autism knowledge, providing more accurate and comprehensive answersAutomatic efficacy evaluations and competitive experiments on system responses indicate our approach significantly improves reliability of autism-related professional knowledge queries.

Ähnliche Arbeiten