Abstract
Recommender systems (RSs) have attracted the attention of many researchers by their applications in various interdisciplinary fields. They help users to overcome the information overload problem and help businessmen to make more profits. Many recommendation algorithms have been discussed and proposed in the literature. Nevertheless, the performance of every single algorithm is limited, and each has its strengths and weaknesses. This paper tries to exploit the power of four RSs, namely, user demographic RS, item demographic RS, user-based and item-based collaborative RSs. Actually, a hybrid recommendation system consisting of two-level hybridization between four individual systems that vary according to the filtering method and the target is proposed. This allows each system to express itself before obtaining the result of the final prediction. Hence, the prediction value will reflect the individual system point of view and consequently different points of view will be available to aggregate the result. For this regard, many schemes for weighing the importance of the prediction result of each system are used. The proposed framework reduces the problem of sparsity and improves the accuracy. Experimental results confirm this finding and show that the proposed framework reduces error and improves coverage compared to traditional ones.