DataTalks #37: Advanced Topics in Large-Scale Recommendation Systems 🪣

Our 37th DataTalks meetup will be hosted by ZipRecuiter at their offices, and will focus on advanced topics in large-scale recommendation systems! 🪣

Location: 18th floor of the southern Hagag Tower, 28 HaArba'a St, Tel Aviv-Yafo
🚘Note: No free parking is available! There are several paid parking lots around; Gindi should be the most inexpensive.

𝗔𝗴𝗲𝗻𝗱𝗮:
🍕 10:00 - 10:20 — Mingling, etc.
🔶 10:20 - 11:10 — Large-Scale Optimization of Matches in Labor Markets
🔷 11:20 - 12:10 — Combating Recommendation Cold Start on a Large Scale

***

Talk #1: Large-Scale Optimization of Matches in Labor Markets

Speaker: Tal Perri, Senior Staff Data & ML Scientist @ ZipRecuiter

Abstract: When considering matching markets (like the labor market), the usual, single-sided, recommender system design provides a suboptimal solution for the goal of maximizing the number of matches in the market. The recently-released paper “Optimizing Rankings for Recommendation in Matching Markets” by Yi Su et al. introduced a new recommendation framework jointly optimizing the rankings for all candidates in the market to explicitly maximize the number of matches. We have lately adapted this framework to optimize a large scale marketplace; namely, the Zip labor marketplace. We have introduced some modifications to the original paper along with a distributed computation for the optimization problem. In this talk we will provide an overview of the basic principles of Yi Su’s framework and discuss our modifications to adapt it to Ziprecruiter’s large scale labor market.

***

Talks #2: Combating Cold Start On a Large Scale — An Evaluation Framework for Cold-Start Techniques in Large-Scale Production Settings

Speaker: Moran Haham, Algorithms Team Leader @ Outbrain

Abstract: Mitigating cold-start situations is a fundamental problem in almost any recommender system. In real-life, large-scale production systems, the challenge of optimizing the cold-start strategy is even more significant. We present an end-to-end framework for evaluating and comparing different cold-start strategies. By applying this framework in Outbrain’s recommender system, we reduced our cold-start costs by half, while supporting both offline and online settings. Our framework solves the pain of benchmarking numerous cold-start techniques using surrogate accuracy metrics on offline datasets – coupled with an extensive, cost-controlled online A/B test. In my talk, I’ll start with a short introduction to the cold-start challenge in recommender systems. Next, I will explain the motivation for a framework for cold-start techniques. I will then describe – step by step – how we used the framework to cut the size of our exploration data by more than 50%.

Bio: Moran Haham is an algorithms manager at Outbrain, with over 10 years of experience in the tech industry. She leads recommendation system research, particularly click-through rate prediction on large scale in production. Moran has a Master’s degree in Information Systems from Ben Gurion University where her thesis focused on using active learning for rating elicitation in recommendation systems.