Hybrid Recommender Systems On Netflix Dataset

The rapid development of the Internet has brought the unprecedented prosperity of the Internet-related services industry,and accumulated cascades of data.Facing vast amount of information,personalized recommendation system can help users quickly find their own interests,save the time cost of users,improve segarch efficiency and service quality,and meet user needs better.Thus,the users' satisfaction and dependence on system will be improved,leading to more users and the growth of interest.Collaborative filtering algorithm has been widely used in recmmender system,but with technology improving and life demand increasing,the requirement for recommender system raises.The accuracy of collaborative filtering algorithm may not be enough.In view of this,it is necessary to carry out research and improvement based on user's neighborhood model and matrix decomposition model.Aiming at this,two aspects of the exploration and research are done as follow:(1)A User-based Hybrid Collaborative Filtering Model improved upon the traditional collaborative filtering algorithm is proposed to solve the problem of low accuracy.The new model consists of two models,one building user-based neighborhood model to calculate the user-project scoring pre-stimate with user-item ratings,while another using users' attributes information to do the same.The hybrid model will mix the results as final scoring value.All recommendations would be done based on scoring value.Experiments on the MovieLens 100k and MovieLens lm datasets show that the new model has a higher accuracy and recall rate than the traditional method,which proves the validity of the new model.(2)A weighted regularized matrix factorization model is constructed to solve the problem of sparse user-rating data in matrix factorization.After shifting user and item orders in the original rating matrix,and then split the matrix into several sub-matrixes,weights are introduced into regularized matrix factorization.Experiment on the Netflix dataset shows that in sparse matrices,the matrix decomposition model with weighting value can reduce the predictive error of the scoring value.