Data for Good Seminars invite leading academics from around the world to share how they are using data to address societal challenges. This seminar is hosted by the DSI Financial and Business Analytics Center.

Guest Speaker

Daniel J. Russo, Philip H. Geier Jr. Associate Professor of Business, Decision, Risk, and Operations, Columbia Business School

Financial and Business Analytics Center Chairs

  • Paul Glasserman, Jack R. Anderson Professor of Business, Decision, Risk, and Operations, Columbia Business School
  • David Yao, Piyasombatkul Family Professor of Industrial Engineering and Operations Research, Columbia Engineering

Hybrid Event Details

Monday, November 14, 2022 (11:00 AM – 12:00 PM ET) – Hybrid

In-Person Location: Northwest Corner Building, 14th Floor (DSI Suite) – 550 W 120th St, New York, NY 10027

This event was NOT recorded.

Talk Information

Optimizing Audio Recommendations for the Long-Term: A Reinforcement Learning Perspective

We study the problem of optimizing a recommender system for outcomes that realize over several weeks or months. We begin by drawing on reinforcement learning to formulate a comprehensive model of users’ recurring relationship with a recommender system.  Challenges of measurement, attribution, and coordination complicate algorithm design. We describe careful modeling — including a  new representation of user state and key conditional independence assumptions — which leads to simple, testable recommender system prototypes. We apply our approach to a podcast recommender system at a large online audio streaming service and demonstrate that purposefully optimizing for long-term outcomes leads to large performance gains over conventional approaches that optimize for short-term proxies. Time permitting, I will also touch on a few other interesting problems at the intersection of recommender systems and reinforcement learning.

*This talk is based on joint work with Lucas Maystre and Yu Zhao, from Spotify.