This article introduces the multi-task learning paradigm adopted by various large-scale video recommender systems. It introduces a general setup for such an MTL-based recommender. It highlights several associated challenges and describes solutions adopted by various state-of-the-art recommenders in the industry.
