Research On Movie Recommender System Based On A Hybrid Recommendation Of SVD And SVM

As the era of Web 2.0 is coming,users' various network information data increases with the days,and information overload becomes more and more serious.For one single user,it's so hard for him to get information he needs quickly from the complex networld;for products providers,how to get all users' information together and dig every user' personal potential demands fastly,and then recommend to them with products those may interest them become one big technical d ifficulty.Individuation recommendation as one effective method and key tool to solve information overload emerges showing good applicable prospects on e-commerce and social med ia industries.Collaborative filtering as one of the earliest and most widely used methods in ind ividuation recommendation makes great success.There are data sparsity,cold start,scalability problem and low accuracy problem however.More and more experts and academicians have been provided many d ifferent algorithms and models to solve the difficulties of traditional collaborative filtering.Hybrid recommendation algorithms become a hot research aspect because hybrid algorithms can relieve the stress of single recommendation method and low efficiency.This article shows one hybrid recommendation algorithm based on singular value decomposition and support vector machine which improved collaborative filtering at some points.The main job is as follows:1.To solve the problem of data sparsity of rating matrix,matrix factorization technique is applied to extract key information and acquire three dense preference singular matrix containing users' preference.Matrix factorization overcame greatly the problem of sparsity;2.As for the scalability problem due to the sharp increase of users and items,singular value decomposition is applied to extract key features of users' rating data and reduce the dimension of singular vector of users or items.Compared with trad itional collaborative filtering,it relieves the calculation stress of similarity matrix and solves scalability problem to some degree.3.To avoid the internal memory loss caused by huge amounts of us ers and items,hybrid recommendation algorithm was proposed only to need to store the singular matrix of users or items,which insures the reduction of singular matrix's dimension,ensures prediction accuracy,and saves more storage.There would be no doubtly great meanings to recommender systems with vast data.4.The research on Movie Lens dataset shows the hybrid algorithm based on SVD and SVM indeed relieve the problems of sparsity,scalability and low accuracy.