Automated health information question-answering reimagined: Medical Relation-Aware Encoding Network

Abstract

This article introduces the Medical Relation-Aware Encoding Network (MeRAEN) to improve medical answer generation (MAG) in automated health information question-answering systems. MeRAEN enhances pre-trained large language models (LLMs) by jointly encoding dialog history and medical knowledge graph. To effectively capture semantic and medical relational information, MeRAEN introduces three innovations: (1) a knowledge association matrix that models word-level medical relations, (2) a medical relation-aware dialog graph that models utterance-level medical relations and (3) a Relational Graph Attention Network with Knowledge (KRGAT). Experiments on real-world medical dialog datasets demonstrate that MeRAEN significantly outperforms state-of-the-art baselines in terms of accuracy, diversity, knowledge utilisation, fluency, patient safety and health outcomes. Ablation studies further verify the effectiveness of each component in MeRAEN. These results underscore MeRAEN’s potential to improve MAG systems and inform the design of domain-adaptive Q&A systems for broader applications beyond healthcare.