Experiences from implementing collaborative filtering in a Web 2.0 application

The goal of this paper is to report our experiences from integrating item-based collaborative filtering into the Web 2.0 site linkfun.net. We discuss the necessary steps to implement the selected Slope One algorithm in our real world application. It was necessary to conduct performance optimization to allow for recommendations without any delays in page generation on our site. Firstly, we significantly reduced the data model by including only items similarities for pairs of items where both items been rated by at least k users. Secondly, we precomputed recommended items for users. By analyzing the empirical results, we found out that user activity increased on the site after introducing the recommender. In addition, users rated recommended videos higher on average than others which indicates that the recommender allowed users to find preferred videos more effectively.