Since 2007, the online shopping destination Gilt has offered daily sales focusing on an ever-broadening range of goods. Originally focused on women’s fashion, we have since expanded to feature men’s fashion, baby and children’s products, home decor, and life experiences. To ensure that we present the most relevant items and sales to our millions of users worldwide, we have employed machine learning techniques that enable us to generate personalized customer experiences via our web and mobile properties and through our daily emails. Given the nature of the flash-sale business, in which products are available for only short periods of time and in low-volume amounts, producing meaningful brand and product recommendations for customers is no simple task. In Gilt’s experience, in order to make the most out of item-based collaborative filtering it is necessary to modify the usual approach. This talk will focus on how to effectively recommend relevant brands when your inventory is constantly changing; and how to scale up as your audience, range of product features and business goals grow and become more diverse. Spark’s machine learning library provides a fast and scalable framework which allows Gilt to perform offline experiments to tune and optimize these models in a flexible manner. This talk will cover some of the algorithmic and evaluation-related options for collaborative filtering, and how Gilt used Spark create an approach that worked for us. It will also offer a practitioner’s view of some of the language features that influence how Spark stores and shares data across the cluster, and how Gilt has made choices involving persistence and broadcasting of data.
Zachary Cohn is a machine learning specialist at Gilt, where he works on developing algorithms and machine learning techniques that enhance the company’s personalization, recommendation and discovery initiatives. Before joining Gilt in May 2014, Zachary belonged to the Machine Learning and Digital Acquisition team at American Express–focusing on web- and display media-based personalization and targeted customer acquisition efforts. He received his PhD in mathematics from Stanford University and his B.S. in Mathematics from the University of Chicago. Outside of work, he enjoys playing chess and bicycling.