Text Classification Algorithm Based On Graph Convolutional Neural Network

Social media has developed into an indispensable communication tool.Text data has been growing exponentially.How to manage and utilize these huge text data has become a huge challenge.The use of machine learning methods for text classification management and mining is an effective natural language processing method,and it has also been a research hotspot in the past20 years.Recently,neural network models have been proven to be able to effectively extract important features from text data.Many scholars use neural network models to extract features of Euclidean spatial data and are widely used in text classification in practical applications.However,most text data in practical application scenarios are generated from non-Euclidean spaces,and traditional neural network models perform poorly when processing these non-Euclidean spatial data.In response to the above findings,this dissertation focuses on improving the classification ability of text classification algorithms,and has made progress in the following two aspects:1.Propose a deep-level graph convolutional neural network model for text classification,and discover the interdependence between nodes through the connection information between each data sample(node)in the graph.In order to improve the effect of obtaining information from words and documents,the feature relationship of text is captured from multiple angles,and a new text graph is constructed according to the relevance of words and the relationship between words and documents.In order to obtain sufficient representation information,the idea of residual network is introduced,and a depth graph residual learning algorithm is designed,which can reduce the risk of gradient disappearance.In addition,the Leaky Re Lu function is used as the activation function,which effectively avoids the possible loss of information during the training process,and ensures the stability of the model on various text classification data sets.2.Designed the graph convolutional neural network algorithm for text classification under the BERT graph attention of transfer learning,which can judge the word vector of the current word according to the current context.First,the BERT model masks the current word,and then learns the current word according to the context,so that a more accurate word representation can be learned.Transfer learning is to enrich the "experience" of the model by learning the knowledge of the "outside",so as to process the current related tasks and obtain more accurate data sample characteristics.Therefore,use multiple data sets to pre-train the BERT and transfer the model.Inputting the samples that need to be trained into the learned BERT and obtain the feature representation of the training samples.and proposed the edge attention mechanism,pay attention to the edges between the samples and the words,focus on the impact of words which related to the sample,and achieve the purpose of improving the classification ability of the text classification algorithm.