[Paper Review] Syndrome-aware Herb Recommendation with Multi-Graph Convolution Network

Yuanyuan Jin, Wei Zhang+3 more|arXiv (Cornell University)|Feb 20, 2020

Traditional Chinese Medicine Studies35 references4 citations

TL;DR

This paper proposes Syndrome-aware Multi-Graph Convolution Network (SMGCN), a novel herb recommendation model that mimics traditional Chinese medicine (TCM) practitioners' syndrome induction process by learning implicit syndrome representations from symptom sets. By leveraging multi-graph convolution networks on symptom-symptom, symptom-herb, and herb-herb graphs, SMGCN captures complex interactions and achieves state-of-the-art performance on a public TCM dataset, significantly outperforming existing methods.

ABSTRACT

Herb recommendation plays a crucial role in the therapeutic process of Traditional Chinese Medicine(TCM), which aims to recommend a set of herbs to treat the symptoms of a patient. While several machine learning methods have been developed for herb recommendation, they are limited in modeling only the interactions between herbs and symptoms, and ignoring the intermediate process of syndrome induction. When performing TCM diagnostics, an experienced doctor typically induces syndromes from the patient's symptoms and then suggests herbs based on the induced syndromes. As such, we believe the induction of syndromes, an overall description of the symptoms, is important for herb recommendation and should be properly handled. However, due to the ambiguity and complexity of syndrome induction, most prescriptions lack the explicit ground truth of syndromes. In this paper, we propose a new method that takes the implicit syndrome induction process into account for herb recommendation. Given a set of symptoms to treat, we aim to generate an overall syndrome representation by effectively fusing the embeddings of all the symptoms in the set, to mimic how a doctor induces the syndromes. Towards symptom embedding learning, we additionally construct a symptom-symptom graph from the input prescriptions for capturing the relations between symptoms; we then build graph convolution networks(GCNs) on both symptom-symptom and symptom-herb graphs to learn symptom embedding. Similarly, we construct a herb-herb graph and build GCNs on both herb-herb and symptom-herb graphs to learn herb embedding, which is finally interacted with the syndrome representation to predict the scores of herbs. In this way, more comprehensive representations can be obtained. We conduct extensive experiments on a public TCM dataset, showing significant improvements over state-of-the-art herb recommendation methods.

Motivation & Objective

• To address the limitation of existing herb recommendation methods that ignore the intermediate syndrome induction step in TCM diagnostics.
• To model the holistic syndrome representation from a set of symptoms, emulating how experienced TCM doctors infer syndromes from symptom patterns.
• To learn comprehensive symptom and herb embeddings by constructing and leveraging multiple heterogeneous graphs (symptom-symptom, symptom-herb, herb-herb).
• To improve herb recommendation performance by integrating syndrome-aware representations with multi-graph GCN-based embeddings.
• To validate the effectiveness of syndrome representation learning and multi-graph architecture in real-world TCM prescription data.

Proposed method

• Construct a symptom-symptom graph from co-occurrence patterns in prescriptions to model symptom relationships.
• Build a graph convolutional network (GCN) on the symptom-symptom graph to learn symptom embeddings.
• Construct a symptom-herb graph and apply GCN to learn herb embeddings while capturing symptom-herb interactions.
• Construct a herb-herb graph and apply GCN to learn herb embeddings based on herb compatibility patterns.
• Fuse multiple symptom embeddings via a multi-layer perceptron (MLP) to generate an overall syndrome representation.
• Interact the learned syndrome representation with herb embeddings to predict herb recommendation scores.

Experimental results

Research questions

• RQ1 Can modeling the implicit syndrome induction process improve herb recommendation performance in TCM?
• RQ2 How effective is the multi-graph GCN architecture in capturing complex interactions among symptoms, herbs, and syndromes?
• RQ3 Does learning syndrome representations from symptom sets lead to better generalization than treating symptoms individually?
• RQ4 How do different graph structures (symptom-symptom, symptom-herb, herb-herb) contribute to the final recommendation performance?
• RQ5 Can the proposed method outperform existing state-of-the-art herb recommendation methods on real-world TCM datasets?

Key findings

• The proposed SMGCN model achieves significant performance improvements over state-of-the-art herb recommendation methods on a public TCM dataset.
• The ablation study confirms that the syndrome representation learning component contributes substantially to the model’s effectiveness.
• The multi-graph GCN architecture, which integrates symptom-symptom, symptom-herb, and herb-herb graphs, yields better performance than models using only a single graph.
• The model demonstrates robustness and generalization capability on unseen symptom sets, validating its mimicry of clinical diagnostic reasoning.
• The use of multi-layer perceptron to aggregate symptom embeddings for syndrome representation is effective in capturing holistic symptom patterns.
• Empirical results show that the proposed method outperforms baseline models across multiple evaluation metrics, including hit rate and mean average precision.