目的 针对现有中医处方普遍存在的挖掘深度不足、适用性差等问题，提出一种新的改良中医处方数据挖掘方法。方法 基于复杂网络分析，融合关联规则与聚类分析，通过挖掘单味中药在数据集中的多种适用病机与各病机下的高关联配伍药物，结合多药物对比，最终实现药物配伍规律挖掘及同效药鉴别分析。为提升方法的准确性与实用性，提出全集提升度指标与重复说者-听者标签传播算法（speaker-listener label propagation algorithm，SLPA）分别作为药物网络边权重与药物类别划分算法。结果 在实证研究中有效分析出《丁甘仁医案》中多种安神药的适用病机与核心配伍的差异；全集提升度指标相比目前应用广泛的关联规则分析提升度指标，修正了在计算高频药物关联度时的异常低值问题；重复SLPA算法优化了SLPA算法存在的多项缺陷，相比传统聚类算法，不仅可实现药物的重叠类别划分，还可导出药物在类中的重要性，进而更容易分析出类的整体属性。结论 相比现有药物配伍规律挖掘方法，该方法可更为准确、客观地挖掘出药物配伍所适用的病机，具有更高的临床指导价值。全集提升度指标可成为提升度指标的有效替代指标，重复SLPA算法可成为传统聚类算法的有效替代算法。

Objective Existing traditional Chinese medicine (TCM) prescription data mining methods generally have problems such as insufficient mining depth and poor applicability to TCM prescription data. To solve these problems, a new method of TCM prescription data mining is proposed. Methods This method is a complex network analysis method. It combines the idea of association rule analysis and cluster analysis. By mining a variety of applicable pathogenesis of single TCM in the data set and the highly associated compatible drugs under each pathogenesis, combined with multi-drug comparison, the drug compatibility rule mining and the identification analysis of the same effect drugs were finally realized. In order to improve the accuracy and practicability of the method, universal set lift index and repeated speaker-listener label propagation (SLPA) algorithm are proposed as the edge weight of the herb network and the herb cluster partition algorithm respectively. Results In the empirical study, this method successfully analyzed the different between applicable pathogenesis and core compatibility of various sedative herbs in Ding Ganren Medical Case Records. Compared with the currently widely used lift index, universal set lift index corrected the abnormally low value problem in calculating the association degree of high-frequency herbs. Repeated SLPA algorithm optimizes many defects of SLPA algorithm, compared with traditional clustering algorithm, it can not only divide herbs into overlapping clusters, but also get the importance of each herb in the clusters. This makes it easier to analyze the overall characteristics of individual clusters. Conclusion Compared with the existing method, this method can more accurately and objectively find out the application conditions of herbal collocations, and has higher clinical guiding value. Universal set lift index can be an effective substitute index for lifting degree index. Repeated SLPA algorithm can be an effective alternative algorithm to traditional clustering algorithm.

R914;R283.21

国家重点研发计划（2023YFC3502900）