Joint Extraction of Interaction Features and Attention Mechanism Fused with Entity-Relations

Joint extraction of entities and relations is an essential task in information extraction.Recently, tagging-based models have gained attention but with poor performance on overlapping triplets, which confronted the issues like cascading errors, information redundancy, and sensitivity to subject extraction.To solve the issue of overlapping triplets, we propose a model combined Gated Attention Unit (GAU) and Multihead Attention Mechanism (MHA) in this paper.This model distinguishes features for entities and relations, and performs entity recognition for each head entity, different head entities are matched to the same or different tail entities.Then specific tail entity tagging is applied to determine all possible relations and corresponding tails.At the same time, in order to better solve the long-distance dependence and complex multiple relations between entities, GAU and MHA are used to enhance the semantic representation, effectively identify entities and relations, and enhance the learning ability and generalization ability of the model.Experiment results on the public datasets DuIE2.0 and CMeIE show that our model performs better than baselines.