Abstract:In order to improve collaborative filtering recommender quality, this paper proposes a hybrid collaborative filtering recommender framework integrated kmeans clustering and supervised feature selection and KDICF algorithm. The algorithm uses supervised feature selection methods and techniques to select the item sets strongly related to the predicted item. The item sets constitutes a lowdimensional useritem rating datasets. The unpredicted rating of target user is predicted by the nearest neighbors’ rating from the lowdimensional dense useritem rating datasets after kmeans clustering. The experimental results show that the hybrid KDICF algorithm has excellent performance.