A collaborative filtering algorithm based on Users´ Partial Similarity

Collaborative filtering is one of the most successful technologies for building recommender systems, and is extensively used in many personalized systems. However, existing collaborative filtering algorithms have been suffering from data sparsity and scalability problems which lead to inaccuracy of recommendation. In this paper, we focus the collaborative filtering problems on two crucial steps: (1) computing neighbor users for active users and (2) missing data prediction algorithm. Consequently, we propose an effective collaborative filtering algorithm based on Users´ Partial Similarity (we call it CFUPS for short). CFUPS´s main idea is that we compute the similarity between users rely on partial items with their common interests, not on all common rated items. And we combine items´ attributes similarity and their ratings similarity properly for computing missing ratings. Theoretically, our method is effective in improving the recommendation precision and withstanding data sparsity. In the meantime, the experiment result shows that our proposed CFUPS algorithm outperforms other existing collaborative filtering approaches.