Section Identification to Improve Information Extraction from Chinese Medical Literature

The Chinese medical literature contains a large amount of knowledge. Reducing the effort needed by medical scholars to extract this knowledge requires a literature analysis to identify the key information in each paper. We argue that identifying the sections of a paper would help us filter noise from the paper and increase the accuracy of extracting the experimental findings. In this research in progress, we consider paper section identification as a sentence classification task and apply Conditional Random Fields (CRFs) to tackle the problem. In our model we combine both lexical and structural features to facilitate section identification. Experiments on a human-curated asthma dataset show that our approach achieves a 10%–20% performance improvement over Support Vector Machines (SVMs), and that use of both bag-of-words features and domain lexicons benefit the task.