Personalized recommendation is the process of ranking a large collection of items based on users’ preferences. Accurate and efficient ranking of items enables providing personalized search results, marketing strategies, e-commerce production suggestions, and entertainment content. In fact, recent estimates show that recommendation systems drive many Internet businesses; estimates show that recommendation systems drove up to 35% of Amazon’s revenue, 75% of movies watched on Netflix, and 60% of videos on Youtube. In addition, Figure 1 illustrates the fraction of cycles devoted to serving personalized recommendation models in Facebook’s datacenter. We find that recommendation accounts for over 72% of all AI inference cycles.
In addition to their importance, modern deep learning solutions for personalized recommendation impose unique compute, memory access, and storage requirements compared to traditional CNN and RNNs. However, despite their importance and unique characteristics, little research attention has been devoted to optimizing recommendation systems.
To address this gap in the research community, we propose a venue to discuss, share, and foster research into personalized recommendation systems.