One of the fundamental methods used in collaborative filtering systems is Correlation based on K-nearest neighborhood. These systems rely on historical rating data and preferences of users and items in order to propose appropriate recommendations for active users. These systems do not often have a complete matrix of input data. This challenge leads to a decrease in the accuracy level of recommendations for new users. The exact matrix completion technique tries to predict unknown values in data matrices. This study is to show how the exact matrix completion can be used as a preprocessing step to tackle the sparseness problem. Compared to application of the sparse data matrix, selection of neighborhood set for active user based on the completed data matrix leads to achieving more similar users. The main advantages of the proposed method are higher prediction accuracy and an explicit model representation. The experiments show significant improvement in prediction accuracy in comparison with other substantial methods.
