结合注意力机制与图卷积网络的汉语复句关系识别  

Chinese Complex Sentence Relation Identification Based on Attention Mechanism and Graph Convolutional Network

作　　者：郑浩 李源[1] 沈威[2] 陈佳杰 ZHENG Hao;LI Yuan;SHEN Wei;CHEN Jiajie(School of Computer Science,Central China Normal University,Wuhan,Hubei 430079,China;Central for Research on Language and Language Education,Central China Normal University,Wuhan,Hubei 430079,China)

机构地区：[1]华中师范大学计算机学院,湖北武汉430079 [2]华中师范大学语言与语言教育研究中心,湖北武汉430079

出　　处：《中文信息学报》2022年第11期60-67,共8页Journal of Chinese Information Processing

基　　金：国家社会科学基金(18BYY174,19BYY092)。

摘　　要：复句的关系识别是为了区分句子语义关系的类别,是自然语言处理(NLP)中必不可少的基础研究任务。现有研究无法使机器在表层判别缺少显式句间连接词句子的语义关系类型。该文将Attention机制与图卷积神经网络(GCN)相结合应用到汉语复句语义关系识别中,通过BERT预训练模型获取单句词向量,输入到Bi-LSTM获取句子位置表示,经Attention机制得到各位置间权重构建图网络以捕获句子间的语义信息,通过图卷积抽取深层的关联信息。该文的方法对缺少显式句间连接词句子的关系识别达到了较好的识别效果,为进一步研究计算机自动分析、识别处理复句的基本方法奠定基础。实验结果表明,在汉语复句语料库(CCCS)和汉语篇章树库(CDTB)数据集上,与先前最好的模型相比,其准确率分别为77.3%和75.7%,提升约1.6%,宏平均F 1值分别为76.2%和74.4%,提升约2.1%,说明了该文方法的有效性。Relation identification of complex sentences is to distinguish the categories of semantic relations of sub-sentences.Focused on the sentences without explicit connectives,this paper applies the Attention mechanism combined with the GCN to classify the semantic relationship of Chinese complex sentences.The Bert based sentence representation formed by word vector is input into the Bi-LSTM.The sentence position representation is obtained and weighted via attention mechanism.A graph network is then constructed to capture the semantic information between sentences.The experiments on the CCCS and CDTB datasets reveal that the proposed method achieves 76.2%and 74.4%F 1 value,respectively,increasing about 2.1%compared with the previous best models.

关 键 词：关系识别 图卷积神经网络 注意力机制 Bi-LSTM 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]