基于图卷积网络的短文本情感分类标注  

SHORT TEXT SENTIMENT CLASSIFICATION BASED ON GRAPH CONVOLUTIONAL NETWORK

作　　者：佘久洲 叶施仁[1] 王晖[1] She Jiuzhou;Ye Shiren;Wang Hui(School of Information Science and Engineering,Changzhou University,Changzhou 213164,Jiangsu,China)

机构地区：[1]常州大学信息科学与工程学院,江苏常州213164

出　　处：《计算机应用与软件》2022年第11期187-193,214,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(61272367)。

摘　　要：针对基于注意力机制的CNN和RNN模型因易忽略句法上关联词之间的依赖关系,而限制短文本情感分类标注的性能这一问题,提出一种基于图卷积网络的短文本情感分类模型。根据句子的依赖树得到词语间的依赖关系,并利用双向长短时记忆网络从文本中提取句子表征;利用图卷积网络结合依赖关系对句子表征进行编码得到节点表征;结合注意力机制,利用节点表征重新分配句子表征的情感权重,并输入到全连接层,通过判别器判定句子的情感极性。在情感分类基准数据集上的实验结果表明,该模型和现有的经典模型相比性能有了一定的提高。Attention mechanism based on CNN and RNN models usually limit the performance of short text sentiment classification because of ignoring the connection between grammatical related words.Aiming at this problem,we propose a short text sentiment classification model based on graph convolutional network.The model obtained the dependency relationship between words among syntactic dependency trees,and utilized the bidirectional long short-term memory to obtain the sentence representation from text.Following the dependency relationship between words,we designed the graph convolution network to encode the sentence representation to get the node representation,where we combined the attention mechanism to redistribute the sentiment weights of the sentence representation.The output representation generated was fed into a softmax layer to calculate the probability scores of sentiment polarity.The experimental results on the benchmark datasets show that the proposed method is competitive with the existing sentiment classification methods.

关 键 词：情感分类 图卷积网络 依赖树 注意力机制 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]