Research And Implementation Of Film Recommendation System Based On Deep Learning And Sentiment Analysis

The rapid development of the Internet makes the network information explosive growth,which leads to the extremely serious problem of information overload.As one of the main information filtering technologies,recommender system faces some problems,such as sparse user rating data,large deviation of user rating and comment sentiment orientation,insufficient deep-level feature mining of users and items,and so on.Therefore,this thesis researches and implements a recommendation system based on deep learning and comment sentiment analysis for film recommendation.This thesis aims to build a personalized movie recommendation model by using deep learning method combined with user reviews,rating data and movie description information.Firstly,the thesis uses crawler technology to obtain data about Douban movies.Then,the polarity of the comment is divided by sentiment analysis technique,and a hybrid film recommendation model is established based on deep learning.Finally,the design and implementation of the movie recommendation system is completed.The specific research contents are as follows:(1)This thesis expounds the background and significance of the research,analyzes the current research situation of sentiment analysis and recommendation algorithm at home and abroad;and studies the traditional collaborative filtering recommendation algorithm and the current main deep learning recommendation techniques,and analyzes their advantages and disadvantages.(2)When analyzing the emotion of movie reviews,aiming at the problems that the static word vector technology such as Word2 vec can’t learn the deep representation of text and solve the polysemy of the word,and the RNN can’t fully exploit the deep semantics of context and has the characteristics of long-term dependence,this thesis proposes a novel Chinese movie review emotion classification model,i.e.,XL-MHA-Bi LSTM(XMB).Firstly,this thesis uses the model XLNet to generate a context-dependent word vector for distributed representation of the information,and then inputs the word vectors into the Bi LSTM network to analyze and calculate the deep semantics of comments.After that,multi-dimensional semantic features are screened by using Multi-head Self Attention mechanism(MHA);next,this thesis uses the sigmoid function to classify the sentiment polarity.Finally,this thesis verifies the validity of the model XMB by experiments.(3)Aiming at the problems that the traditional system filtering recommendation algorithm is restricted by the problem of data sparsity,the movie comment and description information isn’t fully utilized and can’t learn the deep characteristics of the user and films,and various hybrid recommendation algorithms are faced with the problems of improper presentation of auxiliary information.This thesis proposes a hybrid recommendation model,i.e.,XMB-CNN-GRU.Firstly,this thesis uses the XMB model proposed in this thesis to analyses the comment emotion,and fuses the result with rating matrix and film description information.Then,the hidden vectors of users and movies obtained by matrix decomposition are input into CNN to obtain the deep feature representation.Finally,the deep feature representation of the user and the movie are entered into the GRU network to predict the user’s rating of the movie,and the accuracy of the model XMB-CNN-GRU is verified through experiments.(4)A movie recommendation system is designed and implemented by using the recommendation model proposed in this thesis,My SQL database and Django framework.