Research On Commodity Text Classification Based On Graph Neural Network

Since 2016,our e-commerce has been developing rapidly,the size of the Internet users and transaction volume with a high growth rate in the continuous rise.With the rapid development of e-commerce industry,there are so many kinds of commodities that how to realize the automatic classification of commodities and improve the accuracy and efficiency of commodity classification is an urgent problem to be solved.This article carries on the analysis to the commodity name,uses the intelligent method to realize the commodity automatic classification through the commodity name.The text length of commodity name is short,the context data is sparse,and the semantic information is missing.Aiming at this problem,this paper focuses on the text classification algorithm of graph neural network.The specific contents of this paper include:1.This paper studies and analyzes the text data of commodity names,carries on the descriptive analysis to the data,and explores the key information contained in the data.In order to solve the problem of data imbalance,this paper adopts the method of undersampling,and randomly selects the same number of samples from the rich categories as the rare categories,so as to obtain high-quality data for subsequent modeling and analysis.2.Data pre-processing,mainly divided into word segmentation,word stop,text vectorization.Considering that there are plenty of proper nouns in the names of commodities,this paper extends the participle dictionary,which mainly increases the brand-related vocabulary of e-commerce industry,and generates a participle dictionary for e-commerce commodity names,the goal is to achieve better segmentation effect.3.Research the graph convolutional neural network of classifiers,using a textcategorization model that combines large-scale pre-training with direct-learning,and validate the algorithm with 500,000 labeled product names and category relationships on Git Hub.In the experiment,the training set,the verification set and the test set are divided according to the ratio of 3:1:1,the experimental results show that the accuracy of the model is 89.51 percent,and most of the categories can be recognized accurately,which is about 7 percent better than the performance of the text convolutional neural network,compared with other supervised learning methods,the graph convolutional neural network algorithm proposed in this paper has better classification performance.