Collaborative Filtering Algorithm Based On SVD And User Clustering

With the rapid development of Internet and e-commerce technology,the growing transactions can be finished online.Accompanied by the huge amount of data generated,these data is the great wealth for studying user behavior pattern and mining deep information.Meanwhile these data besides the feature of enormousness,there are features of lack data,non-standardized and not conducive to direct analysis.In order to deal with these data,the recommended algorithm has been more and more applied research,in which the collaborative filtering algorithm is the most representative of the study.With the extensive use of the recommended system,the system's shortcomings are constantly exposed,such as data sparse,scalability and other issues.In order to solve the problems in the recommendation system,especially the problem of data sparse and scalability,this paper proposes a recommendation algorithm based on matrix decomposition and clustering.At the same time,in order to consider the influence of the evaluation time of the item on the recommended results,a time influence model is proposed and the model is combined into the scoring prediction algorithm.Experiments show that both algorithms can improve the performance and recommended quality of the proposed algorithm.The main content of this paper is as follows:?1?Aiming at the problem of data sparse and scalability,this paper proposes a collaborative filtering recommendation algorithm based on SVD and user clustering.First,fill the vacancy value in the sparse score matrix.And delete the data of"fat tail effect"to ensure data quality.In calculating the similarity,we consider the influence of the similarity of the implicit information on the ensemble similarity to ensure that the result is more accurate in the similarity calculation.?2?Analyze the impact of time factors on the recommended results.The based on user clustered algorithm according to user characteristics age,gender and occupation to cluster,and then presents a time influence function that can increase the weight of the recent score.In the based on item clustered algorithm,the item is clustered according to the time the project enters the system.Proposed using weight L_u,ito distinguish the different time attributes score forecast.Finally,two kinds of forecasting score methods are combined to consider the method of calculating the score based on the user and the item integration time factor,which makes the score forecast more accurate and convincing.?3?The experimental part verification the performance of the algorithm on the MovieLens dataset.In the experiments based on SVD and user clustering,the difference with other traditional collaborative filtering algorithms is compared and analyzed.It is verified that the algorithm is more efficient in dealing with data sparse and scalability problems.In the experiment of the time factor algorithm,it is verified that the algorithm further improves the accuracy and coverage.