Research On Topic Model Text Classification Based On Graph Convolutional Neural Network

With the rapid development of Internet technology,text information resources are not only explosive growth,but also contain a lot of available resources.Therefore,it is more and more important to classify text data through the establishment of multi-dimensionality,diversity and statistical models of text data through related methods of machine learning,pattern recognition,and deep learning.Text feature extraction is the most important problem to be solved in text classification.A topic model text classification model based on graph convolution neural network is put forward in the paper.This classification model combines two classification algorithms.The first algorithm is mainly to solve the problem that text subject is easy to be confused in the probability based feature extraction method of subject model.The TF-IDF algorithm,which integrates text category information into the traditional LDA topic model,proposes the labeled LDA model to learn text topic features.In the second algorithm,the graph convolution neural network is used to overcome the disadvantage that convolution neural network can only be used for Euclidean structure data to extract features.The specific steps are as follows: first,we use the word co-occurrence rate to learn the relevance between the word items,and use BM25 to learn the relevance between the word and the document,then construct the graph structure vector with the word item,word and document features,and finally input the graph convolution neural network to extract the text feature information.Through the experiment of news text data set,the experimental results show that:The accuracy of text classification based on the thematic model of graph convolution neural network is 76.4%,the recall rate is 75.2%,and the value is 75.8%.Compared with the graph convolution neural network,the accuracy of text classification method is improved by 3%,the recall rate is increased by 3.4%,and the value is increased by3.2%.Compared with the labeled LDA model,the accuracy of text classification is improved by 3.5%,the recall rate is increased by 1%,and the value is increased by2.3%.It is proved that the method proposed in this paper can effectively improve the accuracy of text classification.