A KG-Enhanced Multi-Graph Neural Network for Attentive Herb Recommendation

Area of Science:

• Computational Medicine
• Artificial Intelligence in Healthcare
• Traditional Chinese Medicine (TCM)

Background:

• Traditional Chinese Medicine (TCM) diagnosis involves syndrome induction, a crucial step for health maintenance.
• Existing herb recommenders use TCM prescriptions to induce syndromes but treat all symptoms equally, leading to coarse representations.
• The importance of individual symptoms in TCM diagnosis is not adequately addressed by current methods.

Purpose of the Study:

• To develop an improved syndrome induction model for Traditional Chinese Medicine (TCM).
• To address the limitation of equal symptom weighting in existing TCM diagnostic approaches.
• To enhance the accuracy of herb recommendations by refining syndrome representation.

Main Methods:

• Leveraging an attention mechanism to dynamically weigh the importance of co-occurred symptoms.
• Integrating a Traditional Chinese Medicine (TCM) knowledge graph to enrich input data and improve representation learning.
• Developing a KG-enhanced Multi-Graph Neural Network (MGNN) for attentive propagation of node features and graph structures.

Main Results:

• The proposed model demonstrates superior performance in syndrome induction compared to state-of-the-art methods.
• Attention mechanism effectively discriminates symptom importance, leading to more granular syndrome representations.
• Knowledge graph integration enhances the quality of representation learning in TCM diagnostics.

Conclusions:

• The developed KG-enhanced MGNN model significantly improves Traditional Chinese Medicine (TCM) syndrome induction.
• Attentive symptom weighting and knowledge graph enrichment are key factors for accurate TCM diagnosis and herb recommendation.
• This approach offers a more sophisticated method for analyzing TCM symptom data and generating personalized treatments.