Research On Text Sentiment Analysis Based On Improved LSTM

In recent years,with the increasing development of the Internet and social networking platform,more and more people begin to con:fide in these platforms,this will produce a large number of speak freely,with text information,the text information mining and analysis,can not only understand the user preferences,can also carry out public opinion monitoring according to the information,so the text sentiment analysis gradually has become a hot research topic.This article aims at how to more accurately analyze the problem of emotional categories from texts,and studies and improves the existing LSTM network.The main work content is as follows:Firstly,construct Convolutional Bidirectional Long Short-Term Memory(ConvBiLSTM)model for sentiment analysis task.The network is mainly constructed by adding a convolutional layer in front of the BiLSTM network to extract primary features,reorganize feature maps,and then input them into the BiLSTM network in serialized form to extract advanced features.The English data set used in this paper is the public imdb movie review data set.The Chinese data set is a product review data set involving six fields.The keras framework is used to construct the ConvBiLSTM network.Then the data was randomly generated in accordance with the ratio of 80:20,and the training samples and test samples were generated randomly.The final results were averaged,and the results were compared with the traditional model classification results.The results show that the ConvBiLSTM network achieves better classification results compared to other traditional network models.Second,In the constructed ConvBiLSTM neural network model,attention mechanism was added,and a new network model ACBiLSTM was constructed to give more attention to the keywords in sentences so as to better obtain the emotional information in the sentences so as to realize More accurate emotional classification tasks.The experimental environment is the same as the one in the work.The final experimental results show that the result after adding the Attention mechanism is better than the result without the attention.