基于随机提示的中文法律领域命名实体识别  

Named entity recognition in Chinese legal domains based on random prompts

作　　者：周鹏[1] 何军[1] ZHOU Peng;HE Jun(College of Computer Science,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川大学计算机学院,四川成都610065

出　　处：《计算机工程与设计》2025年第4期1167-1173,共7页Computer Engineering and Design

基　　金：四川省科技重点研发基金项目(2019KJT0015)。

摘　　要：为解决中文法律领域命名实体识别面临的数据集稀缺和通用命名实体识别模型未充分利用BERT文本推理能力的问题,提出一种基于随机提示的命名实体识别方法。设计专用于法律领域的实体类型信息融合层,通过随机融合多角度的实体类型解释信息,结合BERT和BiLSTM,学习文本中融合实体类型解释信息的上下文语义特征。将命名实体识别任务建模为序列标注任务,通过CRF获取序列的标签信息。实验结果表明,该方法在中文法律领域命名实体识别任务中取得了显著的性能提升,F1值达到93.06%。To address the challenges of scarcity in annotated datasets and the underutilization of BERT text inference capabilities in Chinese legal named entity recognition,a named entity recognition method based on random prompts was proposed.A dedicated layer for information fusion of entity types was designed for the legal domain,which incorporated information from various perspectives through the random fusion of entity type interpretation.By combining BERT and BiLSTM,contextual semantic features were learned that incorporated information about the interpretation of entity types in the text.The named entity recognition task was formulated as a sequence labeling problem,and the CRF was employed to obtain label information for the sequence.Experimental results show a significant performance improvement in Chinese legal named entity recognition,achieving an F1 score of 93.06%.

关 键 词：中文法律实体 深度学习 命名实体识别 随机提示 双向长短时记忆网络 序列标注 条件随机场 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]