GAP: Graph-Assisted Prompts for Dialogue-based Medication Recommendation

2025-05-19

AI Summary

To address hallucination in large language models (LLMs) for medication recommendation—caused by cross-turn loss of fine-grained medical information and insufficient domain knowledge in multi-turn doctor-patient dialogues—this paper proposes a graph-augmented prompting framework. First, it extracts medical concepts and their evolving states from the dialogue to construct a patient-centered, dynamic medical concept graph. Second, it aligns this graph with external biomedical knowledge graphs to generate multi-source retrieval-augmented prompts. Third, it introduces a dialogue-state-aware graph neural prompting mechanism to enable interpretable and traceable reasoning. Evaluated on conversational medication recommendation, our method significantly outperforms strong baselines. It further generalizes effectively to dynamic diagnostic consultation, reducing non-factual responses by 32.7% while concurrently improving clinical credibility and recommendation accuracy.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Medication recommendations have become an important task in the healthcare domain, especially in measuring the accuracy and safety of medical dialogue systems (MDS). Different from the recommendation task based on electronic health records (EHRs), dialogue-based medication recommendations require research on the interaction details between patients and doctors, which is crucial but may not exist in EHRs. Recent advancements in large language models (LLM) have extended the medical dialogue domain. These LLMs can interpret patients' intent and provide medical suggestions including medication recommendations, but some challenges are still worth attention. During a multi-turn dialogue, LLMs may ignore the fine-grained medical information or connections across the dialogue turns, which is vital for providing accurate suggestions. Besides, LLMs may generate non-factual responses when there is a lack of domain-specific knowledge, which is more risky in the medical domain. To address these challenges, we propose a extbf{G}raph- extbf{A}ssisted extbf{P}rompts ( extbf{GAP}) framework for dialogue-based medication recommendation. It extracts medical concepts and corresponding states from dialogue to construct an explicitly patient-centric graph, which can describe the neglected but important information. Further, combined with external medical knowledge graphs, GAP can generate abundant queries and prompts, thus retrieving information from multiple sources to reduce the non-factual responses. We evaluate GAP on a dialogue-based medication recommendation dataset and further explore its potential in a more difficult scenario, dynamically diagnostic interviewing. Extensive experiments demonstrate its competitive performance when compared with strong baselines.

Problem

• Addresses fine-grained medical information loss in multi-turn dialogues
• Reduces non-factual LLM responses via external knowledge integration
• Improves dialogue-based medication recommendation accuracy and safety

Innovation

• Constructs patient-centric graph from dialogue
• Combines with external medical knowledge graphs
• Generates queries to retrieve multi-source information