Joint Entity and Relation Extraction Form Medical Information Based on Potential Relation and CasRel
Xiuli Li, Kai Yang · 2025
The extraction of relational facts from unstructured clinical texts is a major challenge in the field of medical information extraction. Existing joint extraction models frequently fail to meet the demands of this task. In this paper, we present CasRel+TextCNN, a new model designed for entity and relation extraction from medical texts. Compared to previous joint extraction models, our approach delivers superior results. Our model builds upon the CasRel framework by integrating a TextCNN module. The TextCNN component identifies potential relationships within the text, which are then incorporated into the CasRel model. This integration helps reduce redundant relation predictions and enhances the overall accuracy of the model. Experimental results demonstrate that our model achieves an F1-score of 51.9% on the CMeIE- V2 dataset, outperforming five strong baseline models. Compared to the original CasRel model, our approach improves precision by 18.7%, recall by 2.9%, and the F1-score by 12.9%. These results indicate that our model can capture richer semantic features and significantly enhance information extraction.