Design And Implementation Of Book Review Emotion Analysis System Based On Bert Model

With the comprehensive popularization and continuous development of the Internet,the ways of exchanging reading insights and experience are also changing and expanding.The rise of various forums and online shopping platforms contain a large number of users' book review text information.In the era of big data,in the face of massive textual information,digging deeply into the information value in the comment texts can not only provide positive feedback for merchants,but also better understand the real needs and needs of consumers,and potential consumers can It can be used for reference and has important theoretical and practical research significance.The traditional affective analysis methods based on affective dictionary and traditional machine learning perform poorly in semantic information feature extraction,which is difficult to deal with the situation that the expression of online book review text is more living and online neologisms are constantly updated.Based on the above background,this thesis designs and implements a book review text emotion analysis system based on Bert model.In coarse-grained emotion analysis,textcnn network is used to accelerate model training to speed up model updating.In fine-grained emotion analysis,multi-layer LSTM and attention mechanism are used to solve the semantic abstraction problem of review text.The main work and contributions of this thesis are as follows:1)Build a book review text dataset.Collect online users' comment text by writing crawler program.After manual screening and text cleaning,use Jieba word segmentation tool to complete word segmentation.In the emotional tagging part,the tagging method is mainly machine tagging,supplemented by manual tagging,and a total of 60000 book review text data sets are finally constructed.2)In view of the excellent performance of Bert model in NLP(natural language processing),but its high computing power and long training time,a scheme of emotional analysis of book review text based on Bert pre training model and neural network finetuning is proposed.The simulation experiment is carried out on the windows platform.Through the experimental verification,compared with the traditional emotion analysis model based on word2 vec word vector,it performs better in accuracy,recall and other indicators;At the same time,compared with the emotion analysis model realized by single Bert model,the training time is shorter and the computing resources are less consumed,so as to determine the feasibility and superiority of the system.3)In the aspect level fine-grained emotion analysis task,aiming at the problem that the book review text semantic abstraction can not effectively extract semantic information features,this thesis proposes an emotion analysis model that integrates Bert model,Bi LSTM network and self attention mechanism.Through experiments,it is verified that the emotion analysis system designed in this thesis has better emotion analysis effect than various emotion analysis models that combine basic neural network and attention.The simulation is carried out under the windows system,and the created book review data set is used for model training.In the coarse-grained emotion analysis algorithm,the accuracy of using Bert model for emotion analysis is improved by about 2%.After connecting the optimized textcnn,the training time of the model is only about 18 minutes,effectively speeding up the training speed of the model.In fine-grained emotion analysis,the use of Bert model can improve the accuracy by 1.31%,and the use of self attention mechanism can improve the accuracy by 0.9%.