结合GCN和注意力机制的文本分类方法研究

Research of Text Categorization Method Combined with GCN and Attention Mechanism

针对传统文本分类方法中需要手动提取特征和分类进而导致分类准确率不高的问题,提出一种结合图卷积神经网络和注意力机制的文本分类方法。方法首先建立整个语料库的大型文本图,然后将文本图的邻接矩阵和特征矩阵输入到图卷积神经网络中,最后网络输出与注意力机制相结合,利用注意力机制中Self-Attention机制的Query矩阵,Key矩阵和Value矩阵计算Attention值,充分学习文本表示,不断调整网络的输出,最终提高文本分类的准确率。在数据集上的仿真结果表明,所提出的方法与传统文本分类方法相比,其准确率较高。

Aiming at the problem that traditional Text Categorization methods need to manually extract features and classifications, which leads to low classification accuracy, a text categorization method combining graph convolutional neural network and attention mechanism is proposed. This method first built a large text graph of the entire corpus, and then input the adjacency matrix and feature matrix of the text graph into the graph convolutional neural network. Finally, the output of the network was combined with the attention mechanism, and the Attention value was calculated by using the Self-Attention mechanism’s Query matrix, Key matrix, and Value matrix, so as to fully learn the text representation, and constantly adjust the output of the network to ultimately improve the accuracy of text categorization. The experimental simulation results on the data set show that the proposed method has higher accuracy than the traditional text categorization method.