Jointly Extract Entities and Their Relations From Biomedical Text

Entity recognition and relation extraction have become an important part of knowledge acquisition, and which have been widely applied in various fields, such as Bioinformatics. However, prior state-of-the-art extraction models heavily rely on the external features obtained from hand-craft or natural language processing (NLP) tools. As a result, the performance of models depends directly on the accuracy of the obtained features. Moreover, current joint extraction approaches cannot effectively tackle the multi-head problem (i.e. an entity is related to multiple entities). In this paper, we firstly present a novel tagging scheme and then propose a joint approach based deep neural network for producing unique tagging sequences. Our approach can not only simultaneously perform entity resolution and relation extraction without any external features, but also effectively solve the multi-head problem. Besides, since arbitrary tokens may provide important cues for two components, we exploit self-attention to explicitly capture long-range dependencies among them and character embeddings to learn the features of lexical morphology, which make our method less susceptible to cascading errors. The results demonstrate that the joint method proposed outperforms the other state-of-the-art joint models. Our work is beneficial for biomedical text mining, and the construction of the biomedical knowledge base.