综合证素分析和“方名-药名”相似度的方剂主药发现算法

Integrating syndrome factor analysis and "Prescription Name-Medicine Name" similarity to mine primary medicines in traditional Chinese medicine

组方规律的研究对于挖掘中医诊治的基本原理,实现中医现代化具有重要意义.本文针对现有算法在发现方剂主药方面存在的不足,提出了将证素与"方名-药名"相似性分析相结合的主药发现算法JPEA(Joint Primary medicine Extraction Algorithm),该算法通过对名医医案临床数据中证素与药物之间的关联分析,并结合方名与药名相似度的计算,来发现方剂中起主要作用的药物.对于证素-药物关联分析,分别设计了基于点互信息的算法、基于贝叶斯的算法和基于MF-ISF(Medicine Frequency-Inverse Syndrome factor Frequency)的算法.实验结果表明,基于点互信息的方法可以达到76.5%的准确率,明显优于文献中已有算法达到的35.8%.同时,实验结果还表明,方名与药名的相似性对于判断方剂的主药具有重要作用.

The research on prescription regularity is important for mining basic principles of therapy and achieving modernization of Traditional Chinese Medicine （TCM）. To overcome shortcomings of existing primary medicine extraction algorithms, Joint Primary medicine Extraction Algorithm （JPEA） which combines the syndrome factors analysis and the similarity of ＂prescription name-medicine name＂ is pro- posed. The algorithm extracts the primary medicines from prescriptions by analyzing the associations be- tween syndrome factors and medicines of clinical data and calculating the similarity of prescription names and medicine names. Based on the pointwise mutual information, the Bayesian method and the MF-ISF （Medicine Frequency-Inverse Syndrome factor Frequency）, three different algorithms are designed for analyzing the associations. Experimental results show that the algorithm based cn pointwise mutual in- formation can achieve 76.5% accuracy, significantly better than the previous algorithms. Meanwhile, the results also indicate that the similarity of prescription names and medicine names plays an important role in the primary medicines extraction.