Large-Scale Real-Time Product Recommendation at Criteo

Abstract

Performance retargeting consists of displaying online advertisements that are personalized according to each user's browsing history. We show close to three billion personalized ads a day, each of them optimized to generate the best post-click sales performance for our clients. Within this time frame, Criteo's recommender system must choose a dozen relevant products from billions of candidates in a few milliseconds. Our main challenge is to balance the amount of data we use with the processing speed and low-latency requirements of a web-scale environment.