Research On Collaborative Filtering Recommendation Algorithm Based On Reviews,Ratings And User Trust

Among numerous personalized recommendation algorithms,collaborative filtering algorithm has attracted a wide range of research and concern since it can use group intelligence to recommend items to users.However,traditional collaborative filtering algorithm affected by the problem of data sparsity frequently,it’s very difficult for traditional collaborative filtering algorithm to make accurate rating predictions and recommendations in the case of lacking rating data.As a real and reliable review can contain rich information of users’ preference and item features,the research work on rating predictions based on review analysis has received more and more attention form researchers 。 In the interest of solving the fake review problems in current recommendation algorithms based on review analysis,and the data sparsity problems in traditional collaborative filtering algorithms,this paper involved the modified user preferences and user trust into traditional collaborative filtering algorithm,and proposed a new user-based collaborative filtering algorithm based on user ratings,reviews and user trust by analyzing the topic distributions and feedback information included in reviews from users.The main work in this paper includes:(1)Introduced the research background of the algorithm proposed in this paper,and the development situation of personalized recommendation technology at home and abroad,emphatically analyzed several mainstream recommendation algorithms,and expounded common experimental methods and evaluation metrics for recommender system.(2)Described Latent Dirichlet Allocation(LDA)Model based on topic analysis,analysed the topic distribution in users’ reviews,and detailed traditional user-based and item-based collaborative filtering algorithms.(3)Proposed the concepts of review diversity、review help and modified user preferences,calculated user trust based on rating data,combined modified user preferences and user trust,and then designed a new user-based collaborative filtering algorithm based on user ratings,reviews and user trust with the basic idea of traditional collaborative filtering algorithms.(4)Experimental validation of this system is undergone on the electronic device reviews dataset of Amazon,and the results of this experiment are compared with other existing algorithms.The experimental results show that,the proposed algorithm in this paper improved the performance of recommender system effectively when compared to existing algorithms based on review topic analysis and traditional collaborative filtering algorithms.To a certain extent,this algorithm also relieved the problems of data sparsity and fake reviews.