Effective association clusters filtering to cold-start recommendations

In this paper, we focus on how to overcome cold-start problem in the traditional research of recommendations system(RS). The popular technique of RS is collaborative filtering(CF). While in real online RS, CF can´t practically solve cold-start problem for the sparsity ratings dataset. In this paper, we propose a novel efficiently association clusters filtering(ACF) algorithm. Considering hybrid approaches, using clustering and also filtering to relieve cold-start problem. ACF algorithm establishes clusters models based on the ratings matrix. We assume the users in the same cluster, they will have the same interests. On the other hand, different users in different clusters present they will have less common interests. The more users ratings for some item in the cluster, can delegate the opinion of the cluster. So we can use the opinion of the cluster to predict the unkowned ratings. Throught the experiments, our method can enlarge the prediction scope and improve the accuracy.