融合多特征和句法引导的中文命名实体识别  

Chinese named entity recognition integrating multiple features and syntax guidance

作　　者：李莉 奚雪峰 盛胜利[4] 崔志明 周悦尧 LI Li;XI Xue-feng;SHENG Sheng-li;CUI Zhi-ming;ZHOU Yue-yao(School of Electronic and Information Engineering,Suzhou University of Science and Technology,Suzhou 215000,China;Suzhou Key Laboratory of Virtual Reality Intelligent Interaction and Application Technology,Suzhou Science and Technology Bureau,Suzhou 215000,China;Suzhou Smart City Research Institute,Suzhou University of Science and Technology,Suzhou 215000,China;School of Computer Science,Texas Institute of Technology,Lubbock 79401,USA)

机构地区：[1]苏州科技大学电子与信息工程学院,江苏苏州215000 [2]苏州市科技局苏州市虚拟现实智能交互及应用重点实验室,江苏苏州215000 [3]苏州科技大学苏州智慧城市研究院,江苏苏州215000 [4]德州理工大学计算机学院,得克萨斯州拉伯克市79401

出　　处：《计算机工程与设计》2024年第11期3448-3456,共9页Computer Engineering and Design

基　　金：国家自然科学基金项目(61876217、62176175);江苏省“六大人才高峰”高层次人才基金项目(XYDXX-086);苏州市科技计划基金项目(SGC2021078)。

摘　　要：针对基于字符的中文命名实体识别模型中所存在一词多义和实体边界潜在词歧义的问题,提出一种融合多层语义特征和句法依存引导的中文NER模型。将句法依存引导的注意力机制与双向长短期记忆网络(BiLSTM)结合,获得字特征向量。通过迭代卷积神经网络(IDCNN)提取汉字独有特征:部首与拼音。采用协同注意力机制对句法依存引导的多种向量进行特征融合。使用CRF层来获得最佳标记序列。在多个公开数据集上的实验结果表明了模型的有效性。Aiming at the issue of polysemy and non-contextualized potential words in a Chinese named entity recognition model based on characters,the model based on multi-layer semantic features and syntactic dependency guidance was proposed.The attention mechanism guided by syntactic dependencies was combined with the bidirectional long short term memory network(BiLSTM)to obtain character feature vectors.The unique features of Chinese character radicals and pinyin were extracted using iterative convolutional neural network(IDCNN).A collaborative attention mechanism was used to fuse features of multiple vectors guided by syntactic dependencies.The optimal labeling sequence was obtained through the CRF layer.The results obtained through experiments indicated that the model is prominent.

关 键 词：中文命名实体识别 多特征融合 句法依存树 BERT 协同注意力机制 一词多义 潜在词歧义 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]