Active Learning to Reduce Cold Start in Recommender Systems

Abstract

Every time a recommender system has a new user, it does not have enough information to generate recommendations with high precision, this is known as cold start. Adapting this problem to a classification problem allow us to apply Active Learning techniques that, as we well see, offer some methods to, given the less possible information about a new user, make right predictions with higher precision than the standard solutions applied in this situation.