PRBJE: A Novel Pre-Relation Bidirectional Joint Extraction Model

Joint extraction of entities and their semantic relationships to form factual triples from unstructured text is a fundamental task in information extraction. However, existing methods either ignore the interactions between relationships and entities or extract the subjects and objects in a fixed sequential order, which makes these approaches too sensitive to the extraction results of subjects. To address the two issues, we propose a pre-relation bidirectional joint extraction model (PRBJE). Specifically, we first detect potential relationships from the sentence to assist in the subsequent extraction of head and tail entities. Then, we extract all possible entity pairs maintained in the text using two complementary directions: extracting the tail entity based on the head entity and vice versa. Finally, we assign relationships to all entity pairs by a biaffine model, thereby forming complete relational triples. Extensive experimental results demonstrate that our model outperforms state-of-the-art methods, with improved F1 scores across four public datasets, particularly achieving gains of 0.9 and 0.7 on NYT and WebNLG, respectively.