Research On Sentiment Classification Of Short Texts Based On Composite Neural Networ

Today,as the Internet has been integrated into every aspect of social life,it has become a normal behavior for people to express their opinions on news events and entertainment gossip,consume on e-commerce apps and comment on commodities or services,and the information contained in them is of potential value to both businesses and customers.Therefore,it is very meaningful to use natural language processing text sentiment analysis technology to mine and analyze these massive unstructured data texts scientifically and reasonably.At present,text sentiment analysis mainly uses correlation methods based on deep learning.However,most of the existing methods have a single input feature representation,which makes the model unable to fully learn semantic information,and The Chinese text needs to be preprocessed before training,which inevitably leads to some inaccurate word segmentation problems in common word segmentation tools.At the same time,the single neural network will lack the expression of sentence system features,the ability to extract deep features and unable to make full use of text feature information.In this paper,a two-channel composite baseline model is proposed.Word vector and char vector in the baseline model as a model of two input channels at the same time,easy to learn more about the semantic information.On each channel,a composite network composed of bidirectional gated recurrent unit network(Bi-GRU)and convolutional neural network(CNN)is used,so that the advantages of Bi-GRU and CNN are complementary,and the attention mechanism is added to focus on more effective features.In addition,attention mechanism was added to focus on more effective features,and the accuracy and F1 values were both higher than those of the control experimental model,about 3% higher than that of the traditional single model,and about 1.5% higher than that of the contrast mixed model on average.Finally,the model is improved based on the framework of the baseline model,and Attention mechanism is added to CNN,which can strengthen the correlation between words and obtain context information.It is used to replace the Bi-GRU of the circular network,and further improve the parallelization and efficiency of the model.The input of the model still retains the two-channel form of word vector and char vector.However,the gating mechanism GTRU added to the model improves the two-channel combination mode,which is more helpful for char vector to assist word vector.Before classification,in order to improve the expression ability of text structure,Capsule Network is used to add routing mechanism for frame to capture location information.Experimental results show that the improved model is more accurate and efficient than the baseline model.