Research on Relationship Extraction Method Based on Medical Examination Reports

With the continuous accumulation of medical and health data and the rapid development of information technology, Medical Examination Reports (MER) have become an important source of health information. To analyze this information in depth, extracting relational triplets from MER texts has become a core step in building large-scale knowledge graphs. However, most methods focus mostly on local features and ignore the global associations of relationship and entity pairs, which limits their ability to capture comprehensive information in relationship extraction. To overcome this limitation, we propose a global feature-based relationship extraction model GBMRE, which not only enhances the expression of local features but also deeply mines the global semantic information by constructing Entity Relationship Feature Table Mining Layer (ERFLayer). In addition, the interaction between features is enhanced by utilizing the Multi-Head Self-Attention, which effectively mitigates the shortcomings of the existing techniques. The experimental results show that this method outperforms the other relational extraction models discussed in this study.