A Research Of Text Sentiment Classification Based On Deep Learning

With the accelerate progress and comprehensive development of Internet technology,all kinds of social networking media such as bamboo shoots after a spring rain emerged,greatly enriched people's interactive information model.People can express their views and opinions on the Internet anytime and anywhere,which makes the data information in the network increasingly large.Because the information convey the user's sentiments and opinions of things in many cases,in order to be able to extract effective emotional information from the massive for relevant applied research,a series of processing and analysis of the text are required.This has produced a hot research focus of the text sentiment classification,which has important influence on the practical application of public opinion monitoring,commodity marketing,financial analysis and so on.At present,the methods of text sentiment classification mainly include dictionary-based methods and machine learning methods.Deep learning as a special machine learning algorithm has also attracted wide attention in the field of Natural Language Processing.In this paper,the crawler program is first designed to collect film review texts from the watercress movie,and manually annotate the emotion category labels of these film review texts.And then studied the application of dictionary method,naive bayes method and the support vector machine method in text sentiment classification task,and then focus on the processing performance of text sentiment classification by convolutional neural network.In the process of studying convolutional neural network applied to text sentiment classification,a convolutional neural network model based on vocabulary feature is constructed.Firstly,the word2 vec tool is used to vectoring the text,and then the feature extraction method and the vocabulary vector dimension are used to explore its influence on the classification performance of the neural network model.It is concluded that choosing appropriate text feature extraction method and lexical vocabulary dimension can improve the accuracy of text sentiment classification to a certain extent.Considering the convolutional neural network model based on vocabulary feature,which ignores the POS features of the text and the semantic relevance between POS.This paper proposes a method that combines the lexical features and POS features,and proposes two WPCNN(Word and POS Convolutional Neural Network)models according to the feature fusion pattern: splice convolution model and independent convolution model,and constructs these two kinds of structural classifiers on Tensorflow.In view of the POS features have context-sensitive,the word2 vec tool is also used to train the POS vector model.In order to verify the feasibility and effectiveness of the proposed two WPCNN models,a relevant research on the positive and negative sentiment classification problems is performed in watercress critique text,and compared with the dictionary method,naive bayes method,support vector machine method and a convolutional neural network model based on vocabulary vectors.The results show that the WPCNN model with the fusion of POS features can learn more semantic information of text,and the classification performance of positive emotional text and negative emotional text are improved on multiple evaluation indicators.