Entity Relationship Extraction for Chinese Electronic Medical Records Based on Tightly Cascaded Binary Pointers

Relationship extraction, as a subtask of information extraction, converts unstructured medical data in electronic medical records into structured data, which lays the foundation for subsequent knowledge graph construction. The existing relationship extraction ignores the potential relationships of similar entities in a single sentence and fails to make full use of the relationships among the closer entities in the text. Thus, we propose a novel tightness mechanism that increases the correlation between head and tail entities to better determine the relationship between similar entities. The experiments of the proposed model on the public dataset demonstrate that the model can extract the triples more accurately and the F1 value is improved by 2.84% over the baseline model.