Cross-Domain Learning Based Traditional Chinese Medicine Medical Record Classification

In Traditional Chinese Medicine(TCM) area, medical records are the objective record of a doctor´s diagnosis and treatment and they are the basis of the TCM development. However, existing medical records of TCM are derived from books, medical cases, Web and most of them lack the categories information. In this paper, we propose a text classification method for the TCM medical record based on cross-domain topic model. First, we transform the physical books into the digital documents, then tokenize and filter the documents with domain lexicons to achieve the significative sequences of words which largely maintain the topics of original documents. Second, we use the cross domain topic model named Topic Relevance Weighting Model(TRWM) to generate the features. Finally, the generated features are leveraged for the medical records classification and compared with the baselines. The experimental results validate the effectiveness of our method.