An empirical study of the recursive input generation algorithm for memory-based collaborative filtering recommender systems

Recommender system research has gained popularity recently because many businesses are willing to pay for a way to predict future user opinions. Such knowledge could simplify decision-making, improve customer satisfaction, and increase sales. We focus on the recommendation accuracy of memory-based collaborative filtering recommender systems and propose a novel input generation algorithm that helps identify a small group of relevant ratings. Any combination algorithm can be used to generate a recommendation from such ratings. We attempt to improve the quality of these ratings through recursive sorting. Finally, we demonstrate the effectiveness of our approach on the Netflix dataset, a popular, large, and extremely sparse collection of movie ratings.