Font Size: a A A

Personalized Recommendation Research Based On Sentiment Analysis

Posted on:2022-12-21Degree:MasterType:Thesis
Country:ChinaCandidate:Y C DaiFull Text:PDF
GTID:2518306779968819Subject:Automation Technology
Abstract/Summary:PDF Full Text Request
With the increasing influence of the Internet on people's lifestyles.People can directly express their thoughts and suggestions on the online platform,and express some opinions on many products which they have used on the E-commerce platform.These review texts contain information about people's emotional preferences,from which valuable information features can be excavated,which will be of great help to enterprises and people.The application fields of natural language processing technology include sentiment analysis,machine translation and speech recognition.Sentiment analysis uses two main methods to process user text information,namely feature extraction and sentiment classification.In the field of recommendation systems,items can be recommended based on users' clear ratings of items,and personalized recommendations can also be made to users through sentiment analysis technology.By mining behavior data and making personalized recommendations for users,it not only improves the experience of users,but also brings benefits to businesses,and further optimizes service measures.Therefore,this thesis selects the e-commerce data in the Amazon platform to classify emotions,and makes personalized recommendation based on emotion analysis module.The main research contents as follows:(1)This thesis designs a BERT-CNN-Bi GRU(BCB)sentiment classification model,which mainly solves the problem of insufficient expression ability of static word vectors and the problem of insufficient ability of traditional sentiment classification model to extract text sentiment characteristics.In the text preprocessing stage,after preprocessing such as deduplication and word segmentation,the model uses the BERT pre-training model to obtain dynamic word vectors,which not only integrates features such as location information and part-of-speech information into the vectors to obtain deeper word feature information,but also alleviates the semantic bias problem of Chinese,making the model have richer semantic features.In the sentiment analysis module,the combined neural network model of Convolutional Neural Network and Bi GRU network is used to extract feature information and sentiment features in user comment text.The CNN network extracts the local features of the text,and the Bi GRU network extracts the contextual semantics of the text.Combining two different features can obtain deep-level information features and improve the performance of sentiment classification,which can make up for the single CNN network or Bi GRU network to extract sentiment features.Finally,these experiments are carried out on the public Amazon e-commerce data.The experimental models results show that the BCB model is better than other comparison models and improves the performance of sentiment classification.(2)This thesis designs a CNN-Bi GRU-FM(CBF)personalized recommendation model.Many traditional recommendation models are based on user explicit ratings of items,and completely ignoring the hidden feature information contained in user's comment texts,which cann't solve the problem of data sparseness.The model uses the sentiment analysis module in the sentiment classification model above in the coding layer to fully mine the user comment text to obtain the feature information of users and products,which can also alleviate the problem of data sparseness.Firstly,the Chinese text is vectorized,then the user characteristics and product characteristics are extracted in the coding layer,and through the feature fusion layer,the vectors of two different feature spaces are fused to sort the features,finally output the user's predicted product score.The experimental results show that this model has higher recommendation accuracy and improved recommendation performance than the set comparison model.
Keywords/Search Tags:Natural language processing, Text sentiment analysis, Deep learning, User preference, personalized recommendation
PDF Full Text Request
Related items