New prediction methods for collaborative filtering

Companies, in particular e-commerce companies, aims to increase customer satisfaction, hence in turn increase their profits, using recommender systems. recommender systems are widely used nowadays and they provide strategic advantages to the companies that use them. these systems consist of different stages. in the first stage, the similarities between the active user and other users are computed using the user-product ratings matrix. then, the neighbors of the active user are found from these similarities. in prediction calculation stage, the similarities computed at the first stage are used to generate the weight vector of the closer neighbors. neighbors affect the prediction value by the corresponding value of the weight vector. in this study, we developed two new methods for the prediction calculation stage which is the last stage of collaborative filtering. the performance of these methods are measured with evaluation metrics used in the literature and compared with other studies in this field.