An effective and scalable algorithm for hybrid recommendation based on Learning To Rank

摘要

Recently, learning to rank in the domain of recommendation has drawn intensive attention. Though many approaches have been proposed, and proved their effectiveness in providing accurate recommendations, they lack emphasis on diversity. However, the predictive accuracy is not enough to judge the performance of a recommended system and diversity has been regarded as a quality dimension for recommendation. In this paper, we propose a formal model based on learning to rank for hybrid recommendation which integrates diversity. We also propose the representation of diversity features by using entropy based on attributes of users and items. Experimental results in the movie domain show the advantages of our proposal in both accuracy and diversity.