Reliability Estimating By Demographic Matrix in Item-based Recommender Systems

Abstract

Nowadays, with the growth of communication between both users and websites, recommender systems have gained significant essential. These systems filter information to find out the user's interests and make personalized recommendations for them. Currently, it is important to provide high-reliability recommendations, because if the recommendations are unreliable, the system may lose the user at the very beginning. In this paper, a Demographic Matrix of users is proposed, then for estimating the reliability of predictions, we combined it with similarity or entropy matrix between items. Finally, we evaluated our approach by comparing it to some other reliability estimation algorithms by MAE (Mean Absolute Error). The slope of a regression line helps to determine how quickly our MAE change by the increase of reliability values, and in this way, we calculated the impact of our method on MAE reduction. The experiments on MovieLens dataset show that the proposed reliability estimation algorithm, due to its massive impact on MAE reduction, is significantly better than other algorithms.