Evaluating Different Strategies to Mitigate the Ramp-up Problem in Recommendation Domains

Abstract

Recommender Systems (RSs) have assumed a prominent role in e-commerce domains, affecting decisively distinct business phases, such as convert new users into customers. The total absence of information about new users is one of the main challenges in this area, and it is known in the literature as Ramp-up Problem. In this scenario, non-personalized strategy are chosen for simplicity, domain independence and effectiveness. State-of-the-art strategies assume that popular items are more likely to represent useful recommendations when the user profile is not known. In contrast, other strategies consider that diversifying recommendations represents potential chances of attracting new users. This work performs an extensive characterization of this problem, in order to contrast the main existing techniques. Our analyses point to a trade-off of popularity and diversity, suggesting that these two dimensions are essential to the Ramp-up problem. However, the main e-commerce systems insist on presenting only strategies that consider accuracy, prioritizing popularity over diversity. The results show that, indeed, both dimensions are relevant to this important scenario in e-commerce.