A Chinese Medical Entity Relation Extraction Method Driven by Label Fusion

The extraction of medical entity relationships is an key step in promoting medical informationization construction aiming at extracting structured triplet information from medical text. A label fusion-driven Chinese medical entity relation extraction framework is proposed to address the issue of insufficient utilization of entity type labels and relation labels in existing methods. Firstly, the entity relationship extraction task is split into four bidirectional named entity recognition tasks, and the labels of each task are replaced with a fusion of head and tail entity type labels and relation labels. Secondly, a triplet construction strategy is designed to maximize the utilization of the triplets extracted bidirectionally. Finally, a triplet bidirectional filtering model is utilized to filter the candidate triplets. The experimental results show that this method has improved the F1 index by 3.01% compared to GPLinker. In addition, this method has also demonstrated excellent performance in complex scenarios such as overlapping relationships, multiple triplets, and cross-sentence triplets in the medical field.