Research On Collaborative Clustering Ensemble Algorithm Based On User Interest Preference Under Social Relationship

The popularity of the Internet and the development of information technology have led to the explosive growth of data,which makes the problem of "information overload" increasingly serious.The recommendation system,which relies on users' historical information to actively provide personalized recommendations for users,effectively alleviates the problem.The core of recommendation system is to accurately mine users' potential interests and preferences through recommendation algorithm to provide users with accurate recommendation.Therefore,how to deeply and accurately mine users' interests and preferences to improve recommendation accuracy has always been a hot issue in the field of recommendation.In the current research,most mainly build different models based on users' historical ratings and social relationship information to mine users' interests and preferences,but one problem is that they ignore the differences in interests and preferences.Therefore,a more accurate recommendation algorithm is proposed through clearly distinguishing users' interests and preferences.In order to improve the accuracy of recommendation and alleviate the problems of sparse data,cold start and poor scalability of traditional collaborative filtering recommendation algorithm,a collaborative clustering ensemble algorithm based on users' interests and preferences under social relationship(SIPCE)is proposed in this thesis.In view of the fact that most of the researches on unsupervised data sets focus on users' interests and preferences for a single item and lack of mining users' overall interests and preferences,SIPCE algorithm first obtain the type information of items through matrix factorization method to obtain the latent feature vector of items in unsupervised data sets and cluster them in order to study users' overall interests and preferences for item types;Secondly,in order to study the differences of user interests and preferences and improve the recommendation performance,SIPCE algorithm constructs users' interest matrix and preference matrix based on item type by clearly distinguishing the differences of interests and preferences and combining their respective characteristics.Two social relationships of expert users and direct trust users are introduced in the construction process for alleviating the problem of data sparsity and cold start,and then users' interests and preferences are calculated by weighted fusion of items according to evaluation subjects.Then,in order to alleviate the problem of reducing the accuracy of recommendation existing in a single clustering algorithm,the clustering ensemble method is used to cluster the users' interests and preferences matrix respectively before calculating the similarity.Finally,the similarity is calculated based on clustering,and the similarity of interest and preference is further fused to find users with similar interest and preference for recommendation,which makes the recommendation results more accurate.The algorithm proposed in this thesis not only solves the problems of traditional collaborative filtering,but also defines users' interests and preferences clearly,and comprehensively considers the influence of interests and preferences on the recommendation system.On two open datasets,numerous experiments have been carried out to analyze the influence of the parameters in the model on the algorithm,and study the different influence of users' interests and preferences on the recommendation system under different evaluation indexes.The effectiveness of clustering ensemble algorithm in improving the accuracy of recommendation algorithm is also considered.The algorithm proposed in this thesis is compared with other classical recommendation algorithms,proving that SIPCE algorithm can further improve the recommendation performance,and also show good results for cold start users.