Entity guided triplet extraction method with additive attention enhancement

Triplet extraction aims to extract structured triplets in the form of (subject, relation, object) from unstructured text. Triplet extraction can be divided into entity guided and relation guided methods according to the extraction order. Entity guided extraction method has become the mainstream method because it is not prone to error transfer problems. However, existing entity guided triplet extraction methods still suffer from issues such as entity pair redundancy and insufficient utilization of entity pair information. Regarding the issue of entity pair redundancy, most researchers choose to alleviate this problem at the entity pair level, but the results are not ideal. Therefore, a extraction framework combining additive attention and pre trained models is proposed to alleviate the redundancy problem of entity pairs from the perspective of the whole triplet. Moreover, to address the underutilization of entity pair information, an entity pair information fusion module is proposed to fully utilize entity pair information to guide the extraction of relations. The method in this paper achieves better-than-baseline results on two mainstream datasets.