Research On Text Sentiment Analysis And Its Application

With the rapid development of Internet,Internet has become the main channel of information dissemination nowadays.People have left a lot of emotional comment on Zhihu,Weibo and TouTiao.If the corresponding emotional information and tendencies be extraced from these reviews data,it is highly condisive to promoting the development of user decision,public opinion monitoring and information prediction,which is of great value in scientific research and practical application.Traditional methods for text sentiment analysis are corpus based,graph based and machine learning methods,all of which rely on manual editing rules.However,as the data volume is continuely in high rise and the way of expression is becoming richer than before,artificial feature based models have been unable to solve these problems effectively,and new methods is urgently needed.Deep learning has made great improvements in model design,training algorithm and other aspects.Making deep learning a breakthrough in the fields of image recognition and text classification.For this reason,this paper based on deep learning method studies text sentiment analysis Specific research work and contributions include:1.We analyze the extraction method of emotional words which is based on BI-LSTM-CRF model,and propose an improved method for BI-LSTM-CRF model.Additionally,we improve the representation and generalization ability of the model by optimizing input sequence and activation function.2.We analyze the ensemble learning method and design a stacking method based ensemble learning framework.effectively avoiding the problem arising from large-scale artificial feature building,and meawhile improving the efficiency of the model ultimately.3.Based on the prediction of public opinion trend,the public opinion trend prediction system is designed and realized.