融合语义和句法依存分析的图卷积新闻文本分类

Graph Convolution Approach to News Text Classification Combing Semantic Relation and Syntactic Dependency

图卷积神经网络GCN已经广泛应用于文本分类任务中,但GCN在文本分类时仅仅根据词语的共现关系来构建文本图,忽略了文本语言本身的规律关系,如语义关系与句法关系,并且GCN不善于提取文本上下文特征和序列特征。针对上述问题,该文提出了一种文本分类模型SEB-GCN,其在文本词共现图的基础上加入了句法文本图与语义文本图,再引入ERNIE和残差双层BiGRU网络来对文本特征进行更深入的学习,从而提高模型的分类效果。实验结果表明,该文提出的SEB-GCN模型在四个新闻数据集上,分类精确度对比其他模型分别提高4.77%、4.4%、4.8%、3.4%、3%,且分类收敛速度也明显快于其他模型。

Graph convolutional neural network(GCN)has been widely used in text classification tasks.,which are mostly built by the co-occurrence relationship of words in text classification.To capture the semantic relationship and syntactic relationship in a text,this paper proposes a text classification model SEB-GCN that introduce syntactic text graph and semantic text graph on the basis of text word co-occurrence graph.It then adopts ERNIE and residual bi-layer BiGRU network to capture text features.Experimental results show that the classification accuracy of the proposed SEB-GCN model is superior to other models on four news datasets.