Reference document-oriented document-level legal relationship extraction model product and method

The invention belongs to the technical field of natural language processing of legal texts, and discloses a judgment document-oriented document-level legal relationship extraction model product and a judgment document-oriented document-level legal relationship extraction method. The model comprises a document embedding layer used for inputting a document marked and inserted with chapter structure information into Lawformer to obtain context embedding representation and attention score of the document; the feature enhancement layer is used for performing deep feature extraction to obtain enhanced document context representation; the entity interaction layer is used for calculating to obtain an enhanced entity pair; the relation classification layer comprises a linear classification layer and an activation function layer, the linear classification layer is used for calculating a probability matrix logits, and the activation function layer is used for calculating logit according to the logits; and obtaining an e