Research And Application Of Movie Recommendation System Based On User Clustering And Time Based Latent Factor Model

With the rapid development of mobile Internet,data information has also ushered in an explosion.It is difficult for people to quickly find what they really need from the massive data information,and they are facing many problems caused by the overload of data information resources.In many audio and video websites,this kind of problem is particularly obvious.It is difficult for users to quickly search and query the movie information they are interested in,which costs a lot of energy.With the development of recommender system,the problem of information overload is gradually alleviated.The purpose of recommender system technology is to actively recommend items of potential interest to designated users.As one of the most widely used algorithms in recommendation technology,collaborative filtering algorithm has exposed more and more shortcomings with the continuous innovation and development of personalized recommendation technology,such as data sparsity,poor scalability and many other problems,The traditional collaborative filtering recommendation algorithm has been unable to meet the application requirements of personalized project recommendation.Aiming at the problems of traditional collaborative filtering algorithm,this thesis proposes a fusion algorithm,which combines user clustering and latent factor model,and proposes R-CTLFM(A Refined Recommendation Algorithm Based on User Clustering and Time Based Latent Factor Model).In order to solve the problem of poor data sparsity and scalability,considering the impact of project evaluation time on the recommendation results,a time function is proposed,which is integrated into the scoring prediction algorithm.The research contents are as follows:(1)Considering the influence of time factor on score prediction,an latent factor model integrating time function is proposed.This thesis analyzes the influence of time factor on user rating,analyzes Ebbinghaus forgetting curve,proposes time exponential function to improve the time weight of user recent rating,and combines the traditional latent factor model to fill the sparse matrix of user rating in all clustering,which effectively alleviates the sparsity of user rating data in clustering,Considering the influence of time factor on user’s interest preference,the experimental parameters of time based latent factor model are determined by experimental comparison.(2)Aiming at the problem of data sparsity and poor scalability,a collaborative filtering recommendation algorithm based on user clustering and latent factor model is proposed.Firstly,aiming at the problem that k-means clustering algorithm is easy to fall into the local optimal solution,combined with the distributed search characteristics of ant colony algorithm,it is improved;Based on the user’s characteristic attributes,the improved k-means clustering algorithm is used to cluster users,which solves the cold start problem of new user logins,and narrows the calculation range of similarity between users to user clusters with the same characteristic attributes.Greatly reduce the time required to calculate user similarity.At the same time,considering the long tail effect,the hot penalty factor is added in the calculation of user similarity to reduce the impact of hot items on user similarity recommendation.Compared with LFM model and user based collaborative filtering algorithm,the RMSE value of R-CTLFM recommendation algorithm is reduced by 2.5% and 4.1% on average,and the accuracy of recommendation is improved.(3)This thesis uses R-CTLFM algorithm to design and implement a personalized movie recommendation management system.The system is based on B/S architecture,using JSP,servlet and other technologies to develop and realize the basic functional interface,such as personalized recommendation of home page,user self-management and administrator background operation.It can realize personalized recommendation according to different user characteristics and scoring operations,and verify the feasibility and practicability of the recommendation algorithm.