Research On Key Technologies Of Sentiment Analysis For Chinese Short Texts

With the rapid development of the Internet,people's daily life is increasingly related to the Internet.People are used to express their opinions,feelings and thoughts on the Internet,which lead to a lot of short text messages appear on the Internet.It is possible to obtain the users' sentimental tendencies by using the natural language sentiment analysis technology to process these short textual information.These sentimental information is of great value for social public opinion analysis,product sales and improvement.This paper takes Chinese short text as the research object,and uses product as the evaluation of the data set to study the sentiment analysis of Chinese short texts based on machine learning and deep learning.The study of short text sentiment analysis which is based on machine learning,in this paper,Naive Bayes and Support Vector Machine algorithms have been used for constructing a sentiment analysis model.In order to improve the performance of the model,two improved methods have been applied to the generation of the text vector.They use the improved chi-square statistic for feature items selection and then weighted to generate a text vector,and use a weighted word vector method to generate a text vector respectively.The performance of each model were compared by experimental results,which showed that both of the two improved methods proposed in this paper have an positive effect on the performance of the sentiment analysis model.The method of using the improved chi-square statistic combined with the weighted text vector generation method has greater effect on the improvement of the accuracy of the sentiment analysis model.The study of deep text sentiment analysis is based on deep learning,this paper uses the attention-based bidirectional long-short-term memory neural network(AM-BLSTM)model to perform text sentiment analysis.In this model,the Long Short-Term Memory(LSTM)solves the problem that when the Recurrent Neural Network(RNN)trains data,the gradient collapse and long-distance dependent.Bi-directional Long Short-Term Memory(BLSTM)solves the problem that the LSTM model can only obtain information from one direction,and the introduction of the Attention Model(AM)makes the AM-BLSTM model able to give more attention to the emotional vocabulary in the sentence,which can more accurately judge the emotional tendency.Compared with the LSTM model and the BLSTM model for short text sentiment analysis,the results show that the proposed AM-BLSTM model can effectively improve the accuracy of sentiment analysis.In addition,some parameters have great influence on the performance of the sentiment analysis model which is based on the machine learning and deep learning,so a series of experiments were conducted to investigate those parameters' influence on the performance of the sentiment analysis model to adjust to performance of the model to the best.Finally,the performance of all the models constructed in this paper,including machine learning and deep learning,were compared.The AM-BLSTM model was found to be the best performer,and the model building system was demonstrated.