A CRF Based Machine Learning Approach for Biomedical Named Entity Recognition

The amount of biomedical textual information available in the web becomes more and more. It is very difficult to extract the right information that users are interested in considering the size of documents in the biomedical literatures and databases. It is nearly impossible for human to process all these data and it is even difficult for computers to extract the information since it is not stored in structured format. Identifying the named entities and classifying them can help in extracting the useful information in the unstructured text documents. This paper presents a new method of utilizing biomedical knowledge by both exact matching of disease dictionary and adding semantic concept feature through UMLS semantic type filtering, in order to improve the human disease named entity recognition by machine learning. By engineering the concept semantic type into feature set, we demonstrate the importance of domain knowledge on machine learning based disease NER. The background knowledge enriches the representation of named entity and helps to disambiguate terms in the context thereby improves the overall NER performance.