基于图卷积网络的服装评价信息分类问题的研究

Research on classification of apparel comment information based on Graph Convolutional Network

随着互联网的快速发展以及电子设备的逐渐普及,越来越多的人选择在网上购物,买家在购买商品之后,可以通过平台提供的评价系统表达自己对服装产品的感受,因此会产生大量的服装评价信息。由于这些评价信息的标签是通过人工选择的,会受到外在因素的影响,所以具有不确定性。这些不确定性产生的误差会影响到平台以及其他用户对服装产品的评判。针对这一问题,本文研究了一种基于图卷积的分类方法,将单词、文档、主题视为节点,三者之间的关系作为边构建大型异构图网络。将该异构图作为图卷积网络模型的输入,并引入了注意力机制,根据不同邻居节点与某一特定节点的关系具有不同的重要程度,构建了关注矩阵。最后对一个公开的服装评价文本进行实验评估以及分析,实验结果表明本方法取得的分类结果优于传统神经网络。

With the rapid development of the Internet and the gradual popularization of electronic devices,more and more people choose to shop online.After buying goods,buyers can provide their own feelings about clothing products through the comment system provided by the platform,which will generate a lot of apparel comment information.Since the labels of these comment information are manually selected and will be affected by external factors,they are uncertain.The errors caused by these uncertainties will affect the judgment of the platform and other users on clothing products.To solve this problem,this paper studies a classification method based on graph convolution,which regards words,documents,and topics as nodes,and the relationship among the three as edges to build a large heterogeneous graph network.The heterogeneous graph is used as the input of the graph convolution network model,and the attention mechanism is introduced.According to the different importance of the relationship among different neighbor nodes and a specific node,the attention matrix is constructed.Finally,an experimental evaluation and analysis of a public clothing evaluation text are carried out.The experimental results show that the classification results obtained by this method are better than traditional neural networks.