The Research On Text Sentiment Analysis Based On Deep Learning

With the rapid development of the Internet, the number of social media and e-commerce platforms increased dramatically. Users from all over world share their comments and sentiments on the Internet become a new tradition. Applying natural language processing technology to analyze the text on the Internet for mining the emotional tendencies has become the main way in the social public opinion monitoring and the after-sale feedback of manufactory. Thus, the study on text sentiment analysis has shown important soc ial significance and commercial value.The existing methods on sentiment analysis are majorly divided into the approach based on sentiment dictionary and the approach based on machine learning. The performance of sentiment dictionary based approach is larg ely dependent on the quality and coverage of sentiment dictionary. As for the machine learning based approach depends on the feature selection and construction. In recent years, deep learning technology obtained great progresses in the field of natural language processing. Thus, this paper investigates the deep learning based text sentiment analysis approach.The main work of this study includes: Firstly, targeting the problem that the recurrent neural network cannot learn long distant dependence informatio n, we substitute hidden layer of the recurrent neural network to the long short-time memory cell for constructing a long short-time memory recurrent neural network model. This model is then applied text sentiment analysis. The evaluations on the dataset of NLPCC2014 Sentiment Classification with Deep Learning Technology Task(NLPCC-SCDL) show that, comparing to the best system in NLPCC-SCDL evaluation, the F1 value of the positive sentiment classification on the Chinese data set is improved for about 0.2% while the F1 value of negative sentiment classification is improved for about 0.6%. As for the classification for negative sentiment on Chinese and the positive sentiment on English, the performance improvements are limited.Secondly, the convolutional neural network(CNN) based sentiment classification model simply uses a fully connected layer for classification and it cannot perform a nonlinear classification efficiently. Target to this problem, this paper designs and implements a sentiment classification model which combines CNN and Support Vector Machine(SVM). In this model, the words sequences of the input samples are replaced by the corresponding pre-training word vector sequences. The convolutional neural network is then utilized to learn feature vector representation corresponding to each input sample. The learned vector representations are fed to a SVM classifier as features for sentiment classification. The experiment results on the NLPCC-SCDL dataset show that the F1 values of the positive and negative sentiment classification on Chinese data set are improved for 1.2% and 1%, respectively, compared with the best system in the NLPCC-SCDL evaluation. Meanwhile, the F1 values of the positive and negative sentiment classification on English dataset are improved for 2.7% and 2.9%, respectively. Based on our knowledge, this model achieved the highest performance on this dataset.This study has shown that the deep learning technology improves the performance of text sentiment analysis effectively. Especiall y, the model combines the CNN and SVM achieved a promising performance.