An improved hidden vector state model approach and its adaptation in extracting protein interaction information from biomedical literature

Large quantity of knowledge, which is important for biological researchers to unveil the mechanism of life, often hides in the literature, such as journal articles, reports, books and so on. Many approaches focusing on extracting information from unstructured text, such as pattern matching, shallow and full parsing, have been proposed especially for biomedical applications. In this paper, we present an information extraction system employing a semantic parser using the Hidden Vector State (HVS) model for protein-protein interactions. We found that it performed better than other established statistical methods and achieved 58.3% and 76.8% in recall and precision respectively. Moreover, the pure data-driven HVS model can be easily adapted to other domains, which is rarely mentioned and possessed by other approaches. Experimental results prove that the model trained on one domain can still generate satisfactory results when shifting to another domain with a small amount of adaptation training data