Replication of Recommender Systems with Impressions

Impressions are a novel data type in Recommender Systems containing the previously-exposed items, i.e., what was shown on-screen. Due to their novelty, the current literature lacks a characterization of impressions, and replications of previous experiments. Also, previous research works have mainly used impressions in industrial contexts or recommender systems competitions, such as the ACM RecSys Challenges. This work is part of an ongoing study about impressions in recommender systems. It presents an evaluation of impressions recommenders on current open datasets, comparing not only the recommendation quality of impressions recommenders against strong baselines, but also determining if previous progress claims can be replicated.