Research On Classification Of Apparel Evaluation Information Based On Graph Convolutional Network

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 evaluation system provided by the platform,which will generate a lot of Clothing evaluation information.Since the labels of these evaluation 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.In order to make these text evaluation information provide reference value for buyers who purchase clothing in the future,and help businesses improve the quality of clothing,this article proposes a clothing evaluation information classification method based on graph convolutional networks.The main content of the research is summarized.For the following three aspects:Firstly,preprocess the clothing evaluation information and construct a heterogeneous graph network.The article analyzes the quality of clothing evaluation information,selects some data,cleans and integrates the data,and finally uses words,documents,and topics as nodes,and the relationship between the three as edges to build a large heterogeneous graph network.One-hot vectors are used to represent the feature values of words and document nodes,and the word frequency distribution learned by the LDA topic model is used to represent the feature values of the topic nodes,and then the heterogeneous graph is used as the input of the graph convolutional network model.Secondly,a graph convolution model applied to clothing evaluation data is proposed.For the first time,the graph convolution model is applied to the classification of clothing evaluation information.Input the adjacency matrix of the constructed text graph and the feature matrix of the node into the graph convolutional neural network to realize the classification of the text.Considering that some nodes carry more important information in the classification process,this paper introduces an attention mechanism for the graph convolution model.According to the different importance of the relationship between different neighbor nodes and a particular node,the attention matrix training model is constructed to achieve better results.According to the graph convolution model proposed in this paper,experiments are also written for verification.A public clothing evaluation text was selected for experimental evaluation and analysis.The experiment first compared the classification results of CNN,RNN and GCN models,proving that our method is superior to the traditional neural network model in performance,and analyzed the reasons.Finally,observe and analyze the experimental results by changing the experimental parameters.Finally,the article also designs a browser-side system for clothing evaluation text classification.The platform administrator can use the system to manage user information,upload evaluation information and train models to obtain classification results,and finally return the results to the browser.Merchants and buyers can search for detailed evaluation information or view the overall evaluation trend of users on the front page category.The system provides a more efficient and convenient evaluation system service for the platform,merchants,and buyers.