An intelligent recommendation system using preference regularization

We have developed a novel intelligent recommendation system for improving rating prediction by collaborative filtering incorporated with preference regularization. One characteristic of the preference regularizer is that it can effectively capture the rating information and ranking information of items, resulting in a decision that is coherent to both rating and ranking of items. Another characteristic of our designed preference regularizer is that it models the difference in the rating to an item between a pair of users in a probabilistic manner. This essentially imposes soft constraint that similar users should have similar rating to an item. We have conducted extensive experiments on real-world datasets to evaluate our framework. We have also compared our framework with other existing works to illustrate the effectiveness of our framework. Experimental results show that our framework achieves a promising prediction performance and outperforms the existing works.