A collaborative filtering recommendation algorithm based on dynamic and reliable neighbors

Abstract

Collaborative filtering algorithm is currently the most widely used and a very efficient technology in personalized recommendation system. To overcome several defects in the research of the traditional Item-based collaborative filtering algorithm, this paper presents a optimized algorithm in two aspects, which are the selection of neighbors and the prediction of ratings. Firstly, different numbers of neighbors for the items and users are dynamically selected according to the similarity threshold, then the reliability of neighbors of both items and users are calculated. Finally, the more reliable neighbors was selected to predict the results. Experimental with MovieLens data set shows that the new algorithm outperforms the traditional Item-based algorithms significantly on accuracy of predictions.