关键词: 复杂网络, 方药超网络, 双权重度中心性（DWDC）, 药对挖掘, 高阶双权重度中心性（H-DWDC）

Abstract: With the in-depth development of complex network research, using complex network theory to model and analyze traditional Chinese medicine prescription data has become an important method for studying the compatibility of traditional Chinese medicine. However, the traditional “medicine-medicine” network based on the “node-edge” structure cannot reflect the higher-order interaction between prescriptions and Chinese herbs, which affects the accuracy and effectiveness of medicine pair mining. To address this issue, this paper proposes a hypergraph-based method for modeling the prescription hyper-network to capture higher-order correlations between prescriptions and Chinese herbs. The prescription hyper-network is constructed by treating prescriptions as hyper-edges and Chinese herbs in prescriptions as nodes. By comparing the statistical properties of the traditional “medicine-medicine” network and the prescription hyper-network, it is found that the prescription hyper-network has a node hyper-degree distribution that conforms more closely to power-law distribution characteristics and exhibits small-world and scale-free properties. This proves the rationality and feasibility of the prescription hyper-network modeling method. Based on the hyper-network of traditional Chinese medicine formulae, a double weight degree centrality (DWDC) index is proposed to identify commonly used herbs in formulae. The impact of hyperedges on node characteristics is considered, and the fragility and robustness of the network are used as a basis. Through comparative experiments with various algorithms, it is demonstrated that the proposed method can more effectively screen out commonly used herbs in the hyper-network of formulae. The DWDC index is further extended to identify commonly used herb pairs in the second order, and a higher-order index H-DWDC is proposed. Experimental results based on a real dataset show that this method can effectively extract commonly used herb pairs from massive data.

Key words: complex network, prescription hyper-network, double weight degree centrality (DWDC), medicine pair mining, high-order double weight degree centrality (H-DWDC)