The Personalized Collaborative Filtering Recimmendation Algorithm Based On User Interest

In the research of personalized recommendation system, many researchers dorecommendation by users’ personal information and user ratings. However, due to users’personal information is not complete and the matrix of user ratings is sparse,recommendation results are not very ideal. To address these issues, a recommendationmodel is given in this paper, which main body is a collaborative filtering algorithm basedon user. According to exploiting the users’ preference for each attribute, we can get thesimilarity of users’ interest. In this way, the similarity can be combined with thecollaborative filtering algorithm which can get the users’ interests from users’ assessment.The main tasks are as follows.First of all, as an intermediate step of data processing in the collaborative filtering,clustering is used to carry out further classification and prediction. The result of clusteringdirectly influences the effect of the recommendation. In order to enhance the accuracy ofclustering, the clustering algorithm is verified in this paper. Fuzzy C-Means clusteringalgorithm is sensitive to the initial points and imprecise to the edge points’ belongingwhich can not reflect the particularity for edge points. What’s more, the standard ofmeasurement, that lies on the distance between the data point and center point of class. Sothe gravity is introduced into the fuzzy c-means algorithm, the class quality and distanceare combined as a clustering metric standards.Secondly, traditional collaborative filtering technology suffers from several issuessuch as the prediction accuracy, the data sparse and so on. To solve these problems, thispaper introduces the user preferences form for users interested in similarity calculation.First of all, according to the size of the information entropy the public property isdeleted.And then a users’ preference table is built according to the users’ assessment toitems’ attribute. At the same time, users’ interest similarity is calculated which can get ahigher similarity set of nearest neighbors. Then the recognition as the weights of similaritymeasure, The users’ assessment similarity is calculated on the basis of the project clusterset. Finally the two methods of similarity measure together, can be used to predict the itemWhich is not assessed. Finally, the detailed experimental process is designed to verify the contents of thispaper. Experimental results show that the proposed approaches for clustering, recognitionand the formation of users’ preference table are feasible and effective, and the analysis ofexperimental results are made.