Study And Implementation Of Movie Recommendation Based On User Sentiment Analysis And Collaborative Filtering

"Recommended" is an active information push way to provide users with system,it is different from the search with specific user needs,so it requires the inference of user interests.Most of the traditional movie recommendations infer user interest based on user rating about the movie,but ignoring the comments information that express user interest and true feelings.In fact,different users given the same rating may not represent their interest in the film are the same.Therefore,this thesis combines users rating information with comments information about films,studies movie recommendation techniques based on user sentiment analysis and the collaborative filtering.The main work is as follows.(1)Crawl and pretreatment of dataset.Through parsing webpage content and address structure of douban movie page,a crawler algorithm is design.The movie dataset and related user ratings and comments information are ontained.The preprocess including deleting and purifying the data with incomplete and unrecognized information are done.The datasets for researching are built for next research work.(2)The fine-grained sentiment analysis based on user comments.Considering the sentiment similarity in user comments,this thesis researches fine-grained sentiment analysis from the users' comments.In detail,different from the traditional sentiment analysis divided by "positive" and "negative" sentiment,here sentiment is divided into "preferences","sad","disgust","fear",and "surprised" more close to the classification of real emotion.This thesis applies the LDA to extract subject from user comments,applies SVM to classify the subjective and objective sentences,and the use naive bayesian classifier and SVM classifier to fine-grained sentiment classification on subjective sentences.This work will establish foundation for emotional similarity calculation and further collaborative filtering recommendation;(3)The film recommendation based on collaborative filtering.Emotional similarity between users is calculated based on fine-grained sentiment analysis results.The similarity is combined with the user rating similarityfor film recommendation based on collaborative filtering.At the same time,by adopting the mechanism of fill the ease of sparse rating matrix,the recommendation quality is improved.(4)Experiment and evaluation.Through the experiments of douban website movie dataset,the results show the advantage of proposed user sentiment analysis with collaborative filtering is recommended in this thesis.