A multi-task learning based approach to biomedical entity relation extraction

Automatic extraction of high-quality biomedical entity relations from biomedical texts plays an important role in biomedical text mining. Currently, existing methods generally focus on training a single task model for a specific task (e.g., drug-drug interaction extraction, protein-protein interaction extraction), ignoring the correlation among multiple tasks. To solve the problem, we used neural network-based multi-task learning method to explore the correlation among multiple biomedical relation extraction tasks. In our study, we constructed a fully-shared model (FSM) and a shared-private model (SPM) and further proposed an attention-based main-auxiliary model (Att-MAM). Experimental results on five public biomedical relation extraction datasets show that the multi-task learning can effectively learn the shared information among multiple tasks and obtain better performance than the single task method.