Multi-Query Perception Model based on Joint Extraction for Entity Relationship Extraction Task

Named Entity Recognition and Relation Extraction are two essential tasks in Natural Language Processing that are widely used in various fields, such as information extraction and knowledge graph construction. However, there are three primary issues associated with these tasks: entity overlap, lack of effective interaction between two tasks, and insufficient utilization of entity semantic information. To address these issues, this paper proposes a multi-query perception model based on the joint encoding approach. The model employs a multi-query perception decoder to achieve interaction between these two tasks and classifies multiple query vectors to solve the entity overlap problem. Furthermore, it applies the Hungarian algorithm to calculate the loss through reasonable correspondence. Experimental comparisons demonstrate that the proposed model effectively addresses the aforementioned problems and achieves state-of-the-art performance on two public datasets.