Research And Improvement Of Personalized Recommendation Algorithm For Sparse Data

With the rapid development of Internet technology and e-commerce system,the information data on the network began to exponential growth.The face of online mass of data,people are increasingly difficult to obtain information of their own interest,this is the "information overload" problem.In order to solve this problem,the recommended system came into being,it can be based on user preferences to provide personalized recommendation services.At present,in a number of personalized recommendation technology,collaborative filtering recommendation technology relies on its unique advantages,in the field of e-commerce has been widely used,but the traditional collaborative filtering algorithm there is cold start,sparse data and scalability problems,How to solve these problems has always been the main topic of the research system.In this thesis,we focus on the traditional collaborative filtering recommendation algorithm,the cooperative filtering algorithm based on singular value decomposition and the collaborative filtering algorithm based on user interest migration,aiming at data sparsity and user interest migration in collaborative filtering technology.Firstly,this thesis introduces the background of the personalized recommendation system,expatiates the basic principle of the collaborative filtering recommendation algorithm and the commonly used similarity measure method,and briefly introduces several commonly used recommendation indexes in the recommendation system.Secondly,according to the problem of user scoring's sparsity,the collaborative filtering algorithm based on singular value decomposition is deeply studied.The implementation steps and problems of the algorithm are introduced.Then the proposed algorithm based on singular value decomposition and project attribute is proposed.Project attribute information improves the reliability of similarity calculation between projects.Experiments on the MovieLens data set are carried out,and the improved recommendation algorithm is compared with the traditional cooperative filtering algorithm and the existed collaborative filtering algorithm based on singular value decomposition.Finally,a collaborative filtering algorithm based on user interest migration is proposed to provide a time weight for each user score in the forecasting stage,which makes the recent user score have more weight for the problem that the user's interest may change with time.Experiments on the MovieLens data set are carried out.The algorithm is compared with the existing time-weighted collaborative filtering algorithm and the collaborative filtering algorithm based on item clustering to verify the effectiveness of the proposed algorithm.