中文摘要：

目的：运用机器学习技术对中医治疗肝病处方中组方规律进行分析,为肝病临床用药以及新药研发提供参考依据。方法：针对某三甲中医院肝病科近2年肝病治疗处方数据,首先运用复杂网络找出药物之间的关联结构,再运用关联规则、聚类分析等无监督机器学习方法,对其进行比较分析,最终得出中医治肝硬化的组方规律。结果：对589首处方,共计257味中药,高频药物组合包括频繁二项集12项、三项集15项以及四项集14项;支持信≥10%、置信度≥90%的关联规则包括＂陈皮,神曲→白术＂,＂猪苓,陈皮→白术＂等34条;通过聚类分析,发现中药主要以5种特征进行归类。机器学习结果与构建的复杂网络结构完全一致。结论：运用机器学习方法进行中医处方数据分析,并与复杂网络方法相结合,以探究中医治疗肝硬化组方规律的方法确实可行,可为临床治疗肝硬化和找寻新方提供线索。

英文摘要：

This study was aimed to use machine learning techniques for the prescription regularity of traditional Chinese medicine （TCM） in the treatment of liver diseases in order to provide a reference basis for clinical treatment as well as research and development of new drugs. According to the prescription data of liver disease treatment of the last two years in the hepatology department of a triple-A TCM hospital, the related structure between drugs was firstly found by the complex structure of drugs. And then, association rule, cluster analysis and other unsupervised machine learning methods were used. The prescription regularity of TCM in the treatment of cirrhosis was received through the comparison and analysis. The results showed that there were 589 prescriptions with 257 types of Chinese medicine herbs. The high frequency drug combination included 2 items of 12, 3 items of 15, 4 items of 14; support 〉 10%, confidence 〉 90% of the association rule include ＂dried tangerine peel, medicated leaven ~ largehead atractylodes rhizome,＂ ＂polyporus umbellatus, dried tangerine peel ~ largehead atractylodes rhizome＂ and other 34; through cluster analysis, it showed that Chinese medicine was mainly classified by 5 characteristics. The machine learning result was the same as the constructed complex network. It was concluded that the combination of complex network and machine learning methods in the exploration of prescription regularity of TCM in the treatment of cirrhosis were feasible. It provided clinical treatment of cirrhosis and clues for finding new prescription in the treatment of cirrhosis.