A Semi-supervised Method for Extracting Multiple Relations of Adverse Drug Events from Biomedical Literature

Relation extraction of adverse drug events is the basis and key part of biomedical text ming. Related researches mainly focus on single relation extraction between drugs and adverse drug reactions and the effect of noise data also cannot be effectively eliminated. Therefore, the content and quality of extraction are difficult to meet the requirements. To tackle this issue, this paper designs a multiple relations extraction task of adverse drug events from biomedical literature. Four types of relations are defined to characterize the whole adverse drug event. A semi-supervised multiple relations extraction method is proposed by combining the bootstrapping algorithm and double-weighted LSTM. Based on the actual data, a series of progressive experiments have been completed. The experimental results show that the proposed method can effectively extract multiple relations of adverse drug events from biomedical literature.