Joint Entity And Relation Extraction Method Fused With Multiple Information

The development of Internet technology has brought massive amounts of unstructured data, extracting entity relation triples from unstructured text is crucial for the construction of large-scale knowledge graph. Most of the existing extraction methods are based on pipeline extraction. But it has the problem of error propagation and the performance of extracting overlapping triples is low. The joint extraction method can solve the above problems, while the existing joint extraction methods ignore the interaction of two subtasks, the enhancement effect of entity type information and the contribution of words under different relations and topics to sentence semantic embedding. In order to solve the above problems, a multi-information fusion entity relation joint extraction model was proposed. The model integrated the predefined entity type description information for head entity recognition, modeled the relationship as a mapping from the beginning entity to the end entity, and used conditional regularization to enhance the sentence representation, and integrated the relationship perception attention mechanism for relationship extraction. Compared with the current SOTA method, the experimental results show that the F1 value is improved by 1%.