广西师范大学学报（自然科学版）2025，Vol.43Issue(3)：1-11,11.DOI:10.16088/j.issn.1001-6600.2024092703
融合边界交互信息的命名实体识别方法
Fusing Boundary Interaction Information for Named Entity Recognition
摘要
Abstract
As a basic task in natural language processing,named entity recognition(NER)can effectively identify and classify named entities in text.Some progress has been made in entity recognition with span-based methods,but the quality differences between candidate spans are often overlooked.To tackle the problem,a named entity recognition method that fuses boundary interaction information is proposed.A boundary interaction module is used to evaluate the semantic associations and interaction strengths between boundaries,and a boundary interaction information matrix is generated.This matrix is used to identify potential semantic connections between boundaries,guiding the model to recognize and mark high-quality candidate spans.Additionally,a multi-scale dilated convolution module is integrated to reduce the impact of non-entity noise by utilizing the semantic relationships between spans.It is demonstrated through experiments that the method achieves F1 scores of 89.78%,87.37%,and 72.10%on the ACE2005 Chinese dataset,ACE2005 English dataset,and Weibo dataset,respectively.These results represent improvements of 0.67,0.95,and 0.69 percentage points over baseline models,validating the effectiveness of the proposed method for named entity recognition.关键词
自然语言处理/命名实体识别/信息抽取/边界交互Key words
natural language processing/named entity recognition/information extraction/boundary interaction分类
信息技术与安全科学引用本文复制引用
何安康,陈艳平,扈应,黄瑞章,秦永彬..融合边界交互信息的命名实体识别方法[J].广西师范大学学报（自然科学版）,2025,43(3):1-11,11.基金项目
贵州省科学技术基金重点项目([2024]003) （[2024]003）
国家重点研发计划(2023YFC3304500) （2023YFC3304500）
国家自然科学基金(62166007) （62166007）
