Integrating divergent models for gene mention tagging

Gene mention tagging is a critical step for biomedical text mining. Only when gene and gene product mentions are correctly identified could other more complex tasks, such as, gene normalization and gene-gene interaction extraction, be performed effectively. In this paper, six divergent models are implemented with different machine learning algorithms and dissimilar feature sets. We integrate these models to further improve the tagging performance. Experiments conducted on the datasets of BioCreative II GM task show that our best performing integration model can achieve an F-score of 87.70%, which outperforms most of the state-of-the-art systems. We also apply CRF++ to see if Kuo et al.´s integration algorithm based on likelihood scores and dictionary-filtering is portable to another CRF package.