Biomedical Named Entity Recognition Based on Feature Selection and Word Representations

As an important task in biomedical text mining, biomedical named entity recognition (Bio-NER) has increasingly attracted researchers attention. Various methods have been employed to solve this problem and achieved desirable results on the annotated datasets. In this work, we focus on the feature set to reduce the training cost by feature selection and template optimization. Also, we integrate three kinds of word representation learnt in unsupervised way to summarize latent features. The experimental results show that feature selection, template optimization and word representation can promote the performance effectively. After post-processing, our methods achieve an F-score of 88.51% and perform better than most of the state-of-the-art systems.