Experience-based Personalized Diversification of Recommendations

Abstract

Accuracy of the recommendations has long been regarded as the primary quality aspect of Recommender Systems (RS), but there's an increasing cognizance that there are other factors such as diversity that users also value. Despite the increased interest of researchers to improve diversification of recommendations, we find that personalization of diversification has been overlooked. As the preference for diversity changes from person-to-person, we propose a personalized diversification technique which is capable of controlling the trade-off between accuracy and diversity, where personalization is achieved by diversifying the recommendation list with more novel items if the user has shown diverse preferences in the past, and diversifying the recommendation list with more relevant items if the user has shown homogeneous preferences in the past. Moreover, we also introduce a novel recommendation technique which uses the past preferences of a user and the ratings of experienced item category experts in recommendation generation process. As post-filtering approaches generate the final diversified recommendation list by selecting items from a list generated from some RS, we use the recommendation technique we propose in order to generate an initial recommendation list with both novel and relevant items to improve the personalized diversification process. Our experiments and evaluation provides evidence to illustrate the properties of proposed techniques and indicate the proposed approach has comparable results to state-of-art techniques. Moreover, unlike other techniques, our approach can promote both novel and relevant items and also make the diversification process personalized.