A Study On Sentiment Analysis Based On Deep Learning

With the gradual popularization of the Internet and high-speed development the network has brought great convenience to people's lives,at the same time,people are also eager to express their opinions and evaluations on the Internet,these content behind the emotional information has a very important significance.In view of the above phenomenon,developers begin to use the method of sentiment analysis to capture the potential emotions contained in the information,the purpose of this thesis is to improve the accuracy of sentiment analysis,to improve and fuse the most widely used models in the research field of sentiment analysis at this stage,so as to obtain a new model of sentiment analysis,and to prove the effectiveness and superiority of the sentiment analysis models presented in this thesis through experiments.The specific research contents are as follows:Firstly,a new CNN-LSTM sentiment analysis model is constructed by fusing Convolutional Neural Network(CNN)with Long Short-Term Memory Network(LSTM).It works by first passing the word embedding as the eigenvector and using the feature extraction operation by the CNN model,and then passing the feature into the LSTM model in the form of serialization to extract the feature second time,in order to improve the accuracy of the sentiment analysis.Using PyTorch deep learning framework to build CNN-LSTM model and experiment,the results are compared with the traditional CNN model and LSTM model.In the experiment,the author also uses the random initialization word embedding to compare with the pre-trained word embedding,and the results show that the CNN-LSTM sentiment analysis model presented in this paper is more effective than the traditional CNN model and the LSTM model after combining the pre-trained word embedding.Using Bidirectional Long Short-Term Memory Network(BiLSTM)instead of Long Short-Term Memory Network and merging with Convolutional Neural Network,the author constructs another new sentiment analysis model--CNN-BiLSTM,in order to extract the affective factors more accurately,and get higher accuracy of sentiment analysis.Based on the same experimental environment and the same combination of the pre-trained word embedding,the experiment compared CNN model,LSTM model,BiLSTM model,CNN-LSTM model and CNN-BiLSTM model.The CNN-BiLSTM model,which combines the pre-trained word embedding,is best in this experiment.