Studies On Recommendation Algorithms Based On Shrinkage Spectrum Clustering

Recommender systems(RS)can help users find relevant information according to past preferences,and has become a powerful tool to overcome the problem of “information overload”.Recommendation technology has been widely used in business,movies,news,video,music and other consumer scenarios and achieved great success.Matrix decomposition with the awareness of grouping has been proposed in recent years by some scholars aimed to improve the accuracy of recommendation.However,an inevitable difficulty of such method is the specific situational problem,that is,how to group users and items according to different recommendation scenarios,such as movies and news recommendation,etc.In this thesis,we propose an approach based on improved spectral clustering to obtain generally groups of users and items.Firstly,we improve the performance of spectral clustering by using James-Stein shrinkage estimator.Then,the improved spectral clustering algorithm is used to cluster both the users and items on the original rating matrix,where we stress that our algorithm has good adaptability to the sample distribution since it does not need any assumptions about the distribution of sample.Moreover,our improved algorithm can obtain more reasonable groups of users and items even if the dataset is high-dimensional and sparse.The simulation results demonstrate that our method can effectively improve the recommendation accuracy.Meanwhile,it can solve the grouping problem in different recommendation scenarios,and its practicability will be greatly promoted.