使用隨機條件域之智慧型病症辨識方法

Abstract

With the advance of technology and the prevalence of Internet access, more and more users attempt to search medical advice on the Internet, and various healthcare websites thus thrive. Users usually seek assistance from those who own similar experiences on healthcare websites. However, there is a great deal of unreliable information without professional endorsement, as the result, users tend to be misled and their conditions may further deteriorate. Even if there are authoritative practitioners involved, they have problem dealing with heavy demand in daily medical advice. Recently, a number of researches explore the intelligent disease inference system, and simply divide it into two parts: medical named entity recognition and disease normalization. This research mainly focuses on symptom named entity recognition. We conduct the experiments using pre-annotated clinical reports released by International Workshop on Semantic Evaluation 2014 Task 7. For each word in the report, we extract features and categorize them into four groups including lexical/morphological, syntactic, semantic, and combinational features, and then utilize machine learning based approach – condition random fields (CRFs) to construct a model that identifies the span of symptom entities in clinical reports. The system performance is evaluated by precision, recall, and f-measure. Our method outperformed some participants in Workshop on Semantic Evaluation 2014 Task 7. Eventually, we analyze the feature influence and key to improve our system in the future.