Text Feature Extraction And Matching Method Based On Graph Convolution Network

Text matching plays a significant role in the research of natural language processing,which mainly studies the matching relationship between texts.It is widely used in information retrieval,search engine,intelligent customer service and other fields.In recent years,with the development of deep learning,text matching technology has also developed rapidly with the help of deep neural network.However,the current technology has some problems in text feature extraction,such as insufficient information extraction,lack of sequence information in graph convolution network and neglect of some text information.Moreover,in the aspect of text interaction,the traditional technology has some problems,such as insufficient interaction between texts,unable to extract deeper interaction information and so on.Aiming at the above problems in the field of text matching,this paper first proposes a feature extraction and text matching method based on graph convolutional network and co attention(GCN-CA).Firstly,this paper constructs the corresponding text graph on the given dataset.The model uses graph convolution network to extract the spatial features of each node in text graph,and constructs a multi-layer collaborative attention network to realize text interaction and matching feature extraction.Finally,GCN-CA model carries out text matching experiments on Quora data set and ant financial data set,which has better performance than the traditional text matching methods.Based on the above research,in order to better carry out text interaction,we propose a deep feature extraction and text matching model based on graph convolutional network and Resnet(GCN-R).The model introduces word level information based on GCN-CA model.According to the different importance of each word in the text,the word weight information is also integrated.In order to solve the problem of gradient disappearance in convolution network,the model uses residual network to mine deeper features in interaction matrix and realize text matching.Finally,higher text matching performance is obtained on ant financial data set and LCQMC data set.GCN-CA model and GCN-R model proposed in this paper solve some problems in text matching,and make innovative improvements compared with traditional methods in text feature extraction,word embedding representation,text deep interactive matching.Moreover,the method in this paper is compared with multiple other methods,which further proves that it is better than those methods.