Handling preferences under uncertainty in recommender systems

While uncertainty can't be ignored in real-world problems, there is almost no research work addressing this issue in the recommender systems framework, especially all that relates to user ratings preferences. Indeed, the subjectivity of user's rating and his/her changing preferences over time, make them subject to uncertainty. Usually, user's imprecise rating for an item (product or service) is time-dependent information and generally provided much later. Meantime the item may change either by degrading or improving its inherent quality. The rating therefore may deviate, since it doesn't describe faithfully the actual current state of the item. This deviation leads to a form of uncertainty on user preferences that we handle in this paper. We show that uncertainty is an ubiquitous aspect in building recommender systems and its taking into account can help predicting the most accurate items by improving their certainty degrees.