Application Study On Text Sentiment Antalysis Based On Deep Learnting Neural Network Framework

With the rapid development of Internet more and more users bring an explosive growth of sentiment data on network platform.Internet users send hundreds of millions of messages every day.A large part of data expresses user's opinions and sentiment tendencies in these massive text messages.Therefore,researches on sentiment analysis have gradually become important targets in the field of natural language processing.With the development of computer technology,the traditional statistical methods are able to satisfy the demand of big data.Therefore,in order to study sentiment analysis of shopping comment data,this thesis makes use of the end-to-end data processing ability and faster computing ability to build deep neural network framework.This thesis makes an in-depth study of the deep neural network model to solve the problems of sentiment analysis.The important research work is as follows:(1)To introduce the traditional research methods about sentiment analysis,this thesis study statistical language model,world embedding and the existing main deep neural network methods on sentiment analysis.(2)According to analysis of a large number of commentary data,it is found that the existing mainstream commentary data form mostly exist in sentence form and have certain length limitation,for example,the maximum number of comment data of Weibo is 140 words.In this thesis,word marking is performed after word segmentation of whole sentence,and whole ID is trained into word vector,which is added into LSTM framework to ensure integrity of data information.And test classification accuracy is more than 91.23%.(3)Aiming at forward dependence of LSTM on data,BI-LSTM model and BI-GRU model are designed to investigate influence of LSTM and GRU models on research of sentiment analysis.This method utilizes the relevant architecture of the deep learning tool Karas,after the optimization of the model,the final results are 92.76%and 91.69%.(4)Aiming at the optimization of the model,this thesis proposes the Attention mechanism.Since the language model of this thesis is a typical sequence to sequence model.The weight of the data context is selected by utilizing Attention matrix,and accuracy of the data features is enhanced,the final result is 97.83%.