Improved Personalized Recommendation based on Causal Association Rule and Collaborative Filtering

Abstract

There are usually limited user evaluation of resources on a recommender system, which caused an extremely sparse user rating matrix, and this greatly reduce the accuracy of personalized recommendation, especially for new users or new items. This paper presents a recommendation method based on rating prediction using causal association rules. First, users and items are mapped into two feature vectors, which would be minded later to get the causal association rules from the perspective of data mining; then based on the casual association rules, the authors create a preference matrix which would predict the rating of the items that users have not rated; finally a nearest neighbor similarity measure method is designed for personalized recommendation. Experiment shows that the algorithm efficiently improves the recommendation comparing to traditional methods.