Abstract: Acupuncture is an important part of Traditional Chinese Medicine（TCM）. Using emerging technologies to mine hidden regularities from acupuncture treatment information can not only help making?greater contributions to the citizens’ health, but also promote the improvement of modern theoretical system of TCM. Choosing acupoints（namely points） is the key to acupuncture treatment, but the research on acupoint selection for acupuncture treatment with modern technology is still in its infancy. To generate acupoints prescription automatically, an algorithm based on K-nearest neighbor method is proposed in this paper. Through analyzing K records which are most similar to the target patient’s record from a patient record set, the algorithm automatically recommends an acupuncture treatment plan for the patient. According to the characteristics of acupuncture clinical data, normalized symptom names, unigrams and bigrams are adopted as features to calculate the record similarity in this paper. An experiment on a clinical acupuncture data set which contains 6267 records is performed to validate the algorithm, and the experimental result shows that using unigrams and bigrams as features are more suitable for the automatic generation of acupoints prescription and its F-measure can achieve respectively 40.30% and 62.71% after deleting and reserving patients’ subsequent visit data.

Key words: acupoints selection, K-nearest neighbor, feature extraction, symptom normalization

摘要： 针灸是中医的重要组成部分，运用新兴技术挖掘大量隐藏在针灸诊疗记录中的规律，既可推动针灸更好地为国民健康服务，又能促进中医现代理论体系的完善。腧穴（即穴位）的选择是针灸治病的关键，而运用现代技术选择腧穴的研究还处于起步阶段。以数据挖掘技术为手段，提出了一种基于K近邻方法的腧穴处方自动生成算法。该算法通过分析病历库中与目标现病史最相似K条病历的穴位配方，来自动给出患者针灸治疗的推荐方案。为更好地计算病历的相似性，根据针灸临床数据的特点分别采用了规范症状、一元字串（unigram）和二元字串（bigram）三种特征。在包含6 267条针灸临床病历的数据集上对算法的有效性进行了验证，实验结果表明使用一元字串和二元字串的特征更适合腧穴处方的自动生成，在删除或保留患者复诊数据这两种情况下F度量值分别可达到40.30%和62.71%。

关键词: 腧穴选择, K近邻, 特征提取, 症状规范