In order to succeed in connecting customers with relevant products and businesses, Facebook’s underlying recommendation system must produce highly relevant recommendations while operating at an astonishing scale. Nikita Lytkin explains how Facebook uses machine learning technologies developed by its ads ranking, applied machine learning, and AI research teams to enable personalized ecommerce that recommends a vast diversity of products to nearly two billion people.
While building this recommendation engine, Facebook tackled a number of machine learning and engineering challenges arising from the unique scale of the problem. Nikita reviews strengths and limitations of some of the commonly adopted approaches to recommendation systems, such as collaborative filtering, with respect to challenges of incorporating customer and product metadata crucial to the personalization task, handling real-world cold-start scenarios where prior customer and product interaction data may be sparse or unavailable, and scalability to large numbers of users and items. Nikita shares some solutions and discusses the benefits gained through state-of-the-art machine learning technologies. He then demonstrates how ecommerce recommendations can be personalized at scale by leveraging neural network model architectures coupled with online retrieval systems that operate on vector embedding representations of items and users.
Nikita Lytkin leads machine learning teams building new monetization products at Facebook. An engineering and data science leader and advisor, previously, Nikita led teams of machine learning engineers and data scientists at LinkedIn working on making the LinkedIn News Feed highly personalized and engaging for over 400 million members and building novel data products empowering educational decision making by prospective college students. Before LinkedIn, Nikita led a team of machine learning experts in computational advertising at Quantcast. The team drove double-digit increases in performance of Quantcast’s online advertising products and company revenue by developing data-driven solutions for ad delivery, fraud detection, and campaign management. Nikita has coauthored over 20 US patent applications and continues to publish in top-tier machine learning and data mining venues. He also advises companies on building data teams and products powered by machine learning and analytics. Nikita holds a PhD in computer science from Rutgers University, where his research focused on machine learning and its applications on textual and financial data.
©2017, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • firstname.lastname@example.org