Research On Text Sentiment Analysis Based On CNN-RNN Deep Learning Model

With the rapid development of computer technology,more and more Internet users express their views on specific things through social platforms.These comments contain many personal sentiment information.Analyzing these contents is of great value to predict the business and social tendency,which also promotes the development of the sentiment analysis research.The main methods of sentiment analysis are as follows: Based on the sentiment lexicon method,this method needs to construct the sentiment lexicon artificially,and the construction of the dictionary has a great influence on classification accuracy.Based on the traditional machine learning method,this method needs to label the corpus manually,then create the features,and finally use the machine learning algorithm to complete the task.This method is easy to implement and has a small computation,but it relies on feature engineering operation which costs a large number of labor cost.In recent years,with the development of deep learning technology,deep learning model is widely used to solve the sentiment analysis.Compared with the former two methods,the deep learning model is better in feature mining and sequence modeling;on the other hand,it can automatically mine the deep features of corpus and achieve good results in the sentiment analysis task.Based on these analyses,this paper designs and implements the deep learning model as follows:Firstly,MCNN-BGRU model is designed,which combines multi-channel convolution neural network and bi-directional gated recurrent unit neural network.The model uses multi-channel convolution neural network for supervised learning extracting the sentiment features with different granularities by using different sizes of filters,to recognize more sentiment feature classification patterns.The bi-directional Gated Recurrent Unit can mine the semantic dependency with a wider span and effectively identify the sentiment polarity of the text.Secondly,due to the different importance of different words in the text to sentiment classification task,this paper introduces attention mechanism to MCNNBGRU.The attention mechanism can help the model adaptively combine the text context information to get the semantic coding with attention probability distribution,and give high value to the key words.Finally,in order to improve the ability of the model to resist gradient dispersion,Maxout neurons are introduced into the training phase of the model to form the ATTMCNN-BGRUM model.The experimental results show that the model has better performance in text sentiment classification task.This paper also analyzes the training process and testing process of the model,and verifies the validity of Maxout neuron.Finally,the paper analyzes the time consumption of different models to illustrate the improvement of training performance.