20 June 2019
• 18:00u | Networking, food and drinks
• 18:30u | Intro by WW
• 18:40u | Duo talk by Michael Skarlinski (Manager, Data Science at WW) and Brian Graham (Data Scientist at WW) - "Leveraging an in-house modeling framework for fun and profit"
A year ago, WW (formerly Weight Watchers) had no formal data science team and no internal tooling for building recommenders or models. We've since scaled up to a team of seven and deployed 6 different models/recommenders serving disparate areas of the business. Our rapid growth has been facilitated through use of an in-house, in-memory DAG-running framework that unified our team's approach to software development, modeling, and recommendation systems. We will walk through our framework's design and highlight how building a framework can help save time, train new data scientists, and facilitate robust model deployments. We'll finish the talk by sharing examples of our recommenders that we've put into production for both social networking media feeds and recipe recommendations.
• 19:10u | Networking and drinks
• 19:30u | Talk by Yuri Brovman (Applied Machine Learning Team Lead at eBay) - "Large Scale Item Recommender System at eBay"
eBay is a global e-commerce marketplace with 1.2 billion items and 179 million buyers. The scale, diversity, and volatility of our inventory makes it challenging to develop algorithms and systems that generate relevant item recommendations that drive conversion. Yuri will present their approach to addressing these challenges with their latest large scale item recommender system. A recommender system has two main components: an ML algorithm and a production engineering system. On the ML side, eBay utilizes a wide range of modelling techniques from collaborative filtering to deep learning to machine learned ranking (MLR). On the engineering side, Yuri will present their production system architecture which enables serving over 1 billion impressions per day. Yuri will discuss both the algorithmic/ML/data science problems they are working on as well as the engineering components of building a production recommender system.