Prediction of User Preference in Recommendation System Using Associative User Clustering and Bayesian Estimated Value

Abstract

The user predicting preference method using a collaborative filtering (CF) does not only reflect any contents about items but also solve the sparsity and first-rater problem. In this paper, we suggest the method of prediction by using associative user clustering and Bayesian estimated value to complement the problems of the current collaborative filtering system. The Representative Attribute-Neighborhood is for an active user to select the nearest neighbors who have similar preference through extracting the representative attributes that most affects the preference. Associative user behavior pattern 3_UB(associative users are composed of 3-users) is clustered according to the genre through Association Rule Hypergraph Partitioning Algorithm, and new users are classified into one of these genres by Naive Bayes classifier. Besides, to get the similarity between users belonged to the classified genre and new users, this paper allows the different estimated values to items which users evaluated through Naive Bayes learning. We evaluate our method on a large CF database of user rating and it significantly outperforms the previous proposed method.