Research On Movie Recommendation Algorithm For Explicit Feedback And Implicit Feedback

The rapid development of big data and over-abundance of information has brought tremendous challenges to information screening.In order to mine different people's different preferences,perspectives,values,and tastes,a good recommendation system has to consider the diversity of preferences and practicality of information.The purpose of the recommendation system is to generate an ordered list of recommendations(or top-N recommendation list)to reflect each user's preferences.Now a lot of work is done based on the user's explicit feedback data,such as movie ratings.However,in actual scenes,there are often not only explicit feedback data but more implicit feedback data,that is data without clear ratings,such as click,watch,and purchase and so on.This article mainly focuses on the movie recommendation system.During the research,we found that there are usually two kinds of data for movie.One is explicit feedback data,which is data that the user has given clear ratings.However,for a movie website,another type of data can be found from the system log of the website,namely,implicit feedback data,which includes the user's click behavior,the user's browsing history,and the user's viewing but not scoring.This article focuses on these two kinds of data separately.For explicit feedback dataInnovation One:A hybrid collaborative filtering framework model is proposed.This paper proposes an optimization algorithm for the accuracy of the recommendation system.The algorithm first uses the user's features to cluster,and then uses a hybrid collaborative filtering framework to train a model for each cluster after the clustering;When using hybrid collaborative filtering,the traditional user-based collaborative filtering recommendation algorithm is improved on the calculation method of user similarity.Experimental results show that the proposed algorithm can improve the quality of recommendation.For explicit/implicit feedback dataInnovation 2:For implicit feedback data,Rendle S proposes a Bayesian personalization ranking(BPR)algorithm.For the problem of slow convergence of BPR algorithm,Rendle S proposes an improved algorithm AOBPR,but this algorithm can not use explicit feedback data.In order to use two types of data at the same time,this paper proposes an algorithm AOBPR_SVD++ that combines the AOBPR model with the most classical SVD++ algorithm based on matrix decomposition.Experiments in two real data sets show that the improved algorithm improves the accuracy of the recommendation to some extent.