Research on Concept Cluster and Data Filling Algorithm for Sparse Data

Abstract

In order to reduce the impact of sparse data on the recommendation quality of the algorithm, this paper proposes a collaborative filtering recommendation algorithm that combines concept clustering and data filling. First, according to the formal background constructed based on the user-item rating matrix and pruning conditions, the object and attribute-induced concept clusters are obtained respectively, after that the target user's nearest neighbor candidate set is determined from the obtained concept clusters to reduce the noise users that affect the similarity measure. Secondly, the rating of unrated items was estimated according to the weights of users' interest in item attributes and the filling rules defined in the paper, then the data of user-item matrix was filled by the pre-estimation to reduce the sparsity of the data set and improve the recommendation accuracy of the algorithm. Finally, a differentiated similarity method is used to measure the similarity between the target user and the user's neighbor set, then predict the user's rating. The comparison experiments of the algorithm in this paper with 8 representative algorithms on two different datasets show that the proposed algorithm has good performance in both prediction and recommendation accuracy.