A Joint Model to Extract Bio-Events from Biomedical Literatures

Event extraction plays an important role in improving complex natural language processing (NLP) applications. The Bio-Event extraction in GE shared task is important to understand biological processes. Although some researches and GE extraction systems have made great progress in biomedical event extraction from literatures, the methods still need to be studied and the performance of existing systems need to be improved. In this study, we propose a joint model based on hybrid kernel to extract biomedical events from literatures which consists of three phases: extracting candidate event pairs, partitioning data into subsets and classifying extracted candidate event pairs. The features used in the classifiers include flat features and semantic path features. We obtain promising results on GE develop data set. Especially the results of simple events are comparable with the state-of-the-art GE extraction systems.