软件导刊2025，Vol.24Issue(5)：46-52,7.DOI:10.11907/rjdk.241268
融合多头注意力与划分过滤网络的医学文本联合抽取模型
Medical Text Joint Extraction Model Integrating Multi-Head Attention and Partition Filtering Network
摘要
Abstract
Medical texts are characterized by knowledge intensity and specialization,and simple neural networks have limited effectiveness in semantic understanding,resulting in poor entity and relationship extraction results.Therefore,a joint extraction model based on fusion atten-tion mechanism and partition filtering network is proposed.This model adopts an Ernie health pre trained model to better understand the se-mantic information of medical texts,and obtains entity features and relationship features through a partition filtering network;Simultaneously adopting a multi head attention mechanism to capture information from different representation spaces,thereby improving extraction accuracy.In addition,the model also uses a fill in method to determine the relationship between entity types and entities,thereby solving the problem of entity overlap and relationship overlap.The experimental results show that the model performs better than the current mainstream joint extrac-tion models on the medical dataset CMeIE.Compared with the NPCTS model,the F1 value of the proposed model increased by 2.3%and the accuracy increased by 3.0%,confirming its effectiveness in medical text information extraction.关键词
医学文本/注意力机制/划分过滤网络/关系抽取/预训练模型Key words
medical text/attention mechanism/partition filtering network/relation extraction/pre-trained model分类
信息技术与安全科学引用本文复制引用
王旨,赵长名,胡中玉,聂堰..融合多头注意力与划分过滤网络的医学文本联合抽取模型[J].软件导刊,2025,24(5):46-52,7.基金项目
四川省科技计划项目(2021YFG0307) （2021YFG0307）
