Research And Application Of Symptom Information Extraction Based On Conditional Random Field

With the rapid developing of information-based society, the "internet plus" thought has been applied into the innovation of everywhere in life. The "internet+ medical" is urgently needed by medical experience in current society. Symptom information extraction is one of the technical supported for the integration of medical information and the drug recommendation system. In this paper, the symptom information extraction was analyzed. Research contents include:1、 Automatic extract method of symptom information based on conditional random fields. Basing on analyzing the text structure and expressing style of drug instructions, extracts key features of symptom information:the features of symptom main body and symptom expressing styles; the extract model of symptom information is built by combining conditional random fields model and others basic text features.2、Symptom information cross matching algorithm which can automatically mark the result based on symptom information text. By analyzing the relevance of automatic remarking result of symptom main body and symptom expressing style and the identified result of symptom information, the cross matching method of symptom information is developed:cross verify the result of symptom main body and symptom expressing styles according to symptom information identified in later period.3、Drugs recommended methods based on drugs adaptive symptom store. The relationships between indications and drugs and between user readme text and recommended drugs are analyzed. According to the TF-IDF weight calculation method, this paper presents a method for the quantification of symptom weight. Based on regarding it as basic combining symptoms and maximum cover rules, ranking algorithm of relevant drugs is deduced.4、Design several group control experiments which researched validity and advancement of extract method of symptom information based on conditional random fields. The experiments indicate that exploded information can be obtained by basing on step -by-step symptom of conditional random fields. Besides, comparing with the symptom information extract method simply based on conditional random model and implicit Markov model, a higher level precision rate and recall rate can be reached in this way.Experimental results show that the method can improve the accuracy rate and recall rate of symptom information extraction based on conditional random field information extraction, and can accurately identify the symptoms and symptoms. Experimental results show that the proposed method is reasonable and can be explained in the paper.