华中科技大学学报（自然科学版）2025，Vol.53Issue(5)：38-43,57,7.DOI:10.13245/j.hust.250016
基于混合关联度的联合实体与关系抽取
Joint entity and relation extraction based on hybrid correlation degree
摘要
Abstract
Aiming at the joint extraction problem of overlapping triples,a cascade relational triple extraction model based on hybrid correlation degree was proposed.In this model,position relation was first introduced to construct a correlation degree encoder,which combined with bidirectional encoder representations from transformers(BERT)to form a dual encoder structure for transforming the input texts into word vectors with contextual semantics and word-sentence correlation information.Then,the word embedding vectors were input to a subject tagging decoder,and the start and end positions of all the possible head entities were marked by the cascade binary tagging tables to obtain the candidate head entity set.Finally,the candidate head entity set fused with hybrid correlation degree was used as the input of a relation-object tagging decoder to identify the corresponding objects and entity relations.Experimental results on Baidu and the New York Times data sets show that compared with the existing models,the proposed model improves in each evaluation index,which verifies the effectiveness of the proposed method.关键词
联合抽取/重叠三元组/混合关联度/级联二进制/位置邻近和重叠算法Key words
joint extraction/overlapping triples/hybrid correlation degree/cascade binary/position adjacency and overlapping algorithm分类
信息技术与安全科学引用本文复制引用
马长林,宋玉婷..基于混合关联度的联合实体与关系抽取[J].华中科技大学学报（自然科学版）,2025,53(5):38-43,57,7.基金项目
国家自然科学基金资助项目(62272189). （62272189）
