Enhancing multi-label disease diagnosis through hypergraph clustering and multi-classification label entropy

Syndrome differentiation is a crucial step in the diagnosis and treatment of Traditional Chinese Medicine (TCM). Ascertaining one or multiple disease types based on the patient’s symptoms falls within the research scope of multi-label learning. Previous studies have addressed this issue through problem transformation or algorithm adaptation, often neglecting the semantic connections between patient’s symptoms and specific disease labels. In this paper, we propose a novel multi-label disease diagnosis method that combines Hypergraph Clustering with multi-classification Label Entropy for Multi-Label Classification (HCLE-MLC) to tackle the problem of TCM syndrome differentiation. We construct a symptom hypergraph based on the connections between symptoms and disease labels, leveraging a clustering-optimized hypergraph attention network to obtain node representations that incorporate label information. This enables symptom nodes with similar properties to tend to cluster into the same group during clustering. Additionally, we develop a multi-class classifier to acquire cluster disease labels, determining the optimal number of clusters based on the output label entropy. Evaluation on two TCM datasets demonstrates that HCLE-MLC outperforms mainstream multi-label learning methods, and offers a degree of interpretability by clarifying the relationship between symptoms and specific disease labels.