Recommendation Algorithm Based On User Clustering And Preferences

With the development of technology and extensive use of Web2.0,the resources on the Internet are growing rapidly.Users can access a variety of resources from the network,but the search process for interesting content becomes inconvenience.The recommendation system can recommend items to users based on user preferences,avoiding search information from large amounts of data.The recommendation system can increase the utilization of network resources and divided into three categories:content-based recommendation system,collaborative recommendation collaboration and hybrid recommendation system.However in practice situation,the number of users and projects is huge and the user rating data is sparse.When constructing user-project rating matrix appear a large number of blank items,causing the problem of data sparse.The sparse problem of rating data will reduce the accuracy of recommendation.Users are not satisfied with the recommended results.To address the problem of extreme sparseness of user rating data,a recommendation algorithm based on user k-means clustering and preferences is proposed.The k-means algorithm is easy to fall into the local optimal solution when dealing with non-convex functions.The clusters of k-means are spherical in shape.The algorithm does not consider the user preferences change over time.Based on the above two points,this paper proposes the following improvements:Spectral clustering can cluster the data in any shape of the sample spaces.Spectral clustering algorithm based on information entropy can avoid the scale parameters require manual control problem,so using spectral clustering algorithm based on information entropy to cluster users.Considering the influence of time on user preferences,so introduce the time influence factor.The definition of time influence factor refers to the Ebbinghaus forgetting curve.The algorithm uses the time influence factor to optimize the user-project rating matrix.The improved recommendation algorithm based on user clustering and preferences can deal with multiple types of data and get the global optimal solution.The algorithm can improve the accuracy of recommendation by introducing time influence factor which can optimize the user-project rating matrix.