Research On Chinese Short-Text Sentiment Multiclass Classification Based On Deep Learning

With the rapid development of the Internet industry,the traditional way of communication between people has changed,and the short text data is accumulating in large numbers.The analysis of the big data is extremely urgent.The sentiment classification of short text is an important part of the current research filed and an important aspect of natural language processing research.What we need is not just for the short text briefly analysis of the positive and negative to the sentiment,more categories,deeper levels,more accurate and detailed emotion is our current research purpose.Machine learning models are mainly models of traditional classification models,such as Support Vector Machine(SVM),Bayes classifier,Decision tree and so on.In recent years,the deep learning model has surpassed the machine learning model with its unique advantages.This paper proposes a VC corpus composition method,a WCMG corpus composition method and a new deep learning fusion model.For VC corpus composition method,through VC binomial generating method,we make small and unbalanced sample categories corpus constitute new corpus,in order to realize the expansion of small corpus and the balance of the sample classification.For WCMG corpus composition method,we merge Word2vec word vector and Glove word vector which both have been processed by VC method in a new way.It can constitute a new tensor,complement each other's advantages and extract data feature better.In the new deep learning fusion model,after reconstructing and analyzing experimental of a lot of deep learning classification models that have been proposed,we try to do a certain degree of transformable experiments and propose a unique model fusion method.From the comparative analysis of numerous experimental results,the VC corpus composition method can significantly improve the accuracy of the model.However,the WCMG corpus composition method and the new deep learning fusion model slightly improve the accuracy.There still have some room for improvement.It follows that comparing with traditional way of dealing with corpus and traditional deep learning model,the VC corpus composition method,the WCMG corpus composition method and the new deep learning fusion model have stronger feature extraction ability and model generalization.There is no doubt that they can improve the accuracy of the short text sentiment classification.