Improved Collaborative Filtering Algorithm Based on Multifactor Fusion and User Features Clustering

Nowadays, collaborative filtering has become a core technology of personalized recommender systems, but the data sparsity and scalability problems seriously affected the accuracy of similarity computation. To alleviate these problems, a novel effective similarity calculation algorithm combined with user features clustering was proposed in this paper. The newly users similarities calculation model integrated the improved user rating similarity and items category preference similarity. Then, considering the K-means clustering algorithm, we clustered the users based on the user attributes through Euclidean distance so as to the nearest neighbors could be obtained in proper clusters. Experiments on MovieLens dataset are implemented. The experiments show that, compared with other improved algorithms, the method can effectively improve the prediction accuracy and quality of recommendation system.