Joint Relation Extraction Model Based on Potential Relation and Entity Boundary Information

Extracting entities and relation from unstructured text is a crucial step in building knowledge graphs. Existing methods have achieved certain effectiveness, but issues such as entity nesting and relation redundancy still persist. Regarding the mentioned problems, a method of Potential Relation and Entity Boundary Information (PREBI) is proposed. At the start, the model uses a pre-trained model for representing the encoded input text, then predicts the relation present in the sentence through a potential relation extraction module, incorporating entity boundary information to jointly model and generate relation triplets. Experiments conducted on the DuIE2.0 dataset demonstrate outstanding performance on Chinese data, with an F1 score of 76.41%.