An Entity Enhancement-Based Approach for Joint Extraction of Entity Relationships in Medical Texts

Medical text is characterized by dense terminology, high context dependency, scarce and complex data, which makes the task of joint extraction of entity relationships in medical text challenging. In order to solve this problem, a joint extraction method based on entity reinforcement is proposed for the extraction of medical text entities. First, the BERT pre-trained model was used to extract the representation vectors of medical text; then a global entity strengthening layer was built to obtain the global vector information of the text by constructing a head-to-tail entity acquirer; then a local entity strengthening layer based on an inflated convolutional network was constructed to capture the local contextual information of the text; and finally a five-type corner labeling strategy was adopted to decode the pairs of entities and their relationships by constructing a global two-dimensional matrix. The experiments show that the joint entity-relationship extraction model constructed in this paper can improve the extraction effect on medical text by 1.62% compared to the mainstream methods, and also performs well in complex scenes.