Utilizing semantically enhanced self-supervised graph convolution and multi-head attention fusion for herb...

Area of Science:

• Integrative and Complementary Medicine
• Artificial Intelligence in Healthcare
• Pharmacology

Background:

• Traditional Chinese Medicine (TCM) is a recognized natural therapy.
• Herb recommendation systems are crucial for TCM application but face challenges.
• Existing systems suffer from data sparsity and poor representation of symptoms and herb characteristics.

Purpose of the Study:

• To introduce a novel herb recommendation approach, BSGAM.
• To address limitations of existing TCM recommendation systems.
• To improve the accuracy and effectiveness of herb recommendations.

Main Methods:

• Utilizing BERT for efficient entity embedding.
• Employing a residual graph convolution network and self-supervised learning for feature optimization.
• Integrating a multi-head attention mechanism for feature fusion and recommendation.

Main Results:

• BSGAM demonstrated significant improvements over baseline methods.
• Achieved 6.80% increase in F1-Score@5, 7.46% in F1-Score@10, and 6.60% in F1-Score@20.
• Validated effectiveness on a public TCM prescription dataset.

Conclusions:

• The BSGAM approach effectively enhances herb recommendation accuracy.
• Semantically enhanced self-supervised graph convolution and multi-head attention fusion show promise for TCM.
• This method offers a more robust solution for personalized herbal medicine recommendations.