Program (October 14 2024)
Program (October 14 2024)
Embedded Files
Invited Talks
Invited Talks
Talk 1: “Optimizing for a broader world of entertainment”
Gary Tang (Netflix)
Abstract:
Our mission at Netflix is to Entertain the World. Beyond recommending something to play in the moment, our goal is to consistently provide a joyful experience and create satisfaction over the long-term. Historically, that choice has been over which show or movie to recommend. Today, Netflix has more ways than ever to bring joy to our members. In addition to movies and shows on demand, we offer games and live events. Recommender systems face several challenges in properly recommending new content categories. Firstly, users may disregard such recommendations due to lack of familiarity. Secondly, new content categories are commonly mismatched with users habitualized around visiting Netflix to find their next bingeable movie and show and not, for example, inclined to play a game. Finally, the recommender system may under expose the content due to low engagement stemming from the two preceding challenges. In the initial stages, it often appears as if the new initiatives have a lower expected long-term value relative to existing offerings. Optimizing purely on existing understanding of long term member satisfaction could stifle changing user attitudes and responses. In this transitory state, a practical approach is to design recommenders to maximize the multiple objectives of long term user satisfaction and potential of new products. In this talk, we will discuss our approach to this problem.
Talk 2: “Strategy meets Relevance: The Art of Combining Strategy and Relevance for Content at Spotify”
Tonia Danylenko & Roberto Sanchis Ojeda (Spotify)
Abstract:
In this talk, we'll explore Spotify's journey in building a strategic content platform, tackling challenges like cold-starting new content, balancing reach and precision, and optimizing for long-term impact. We'll dive into how we surface relevant content to drive engagement and enhance user experience, all while navigating the trade-offs between strategy and relevance. Join us for insights into the art of content discovery at Spotify.
Gary Tang (Netflix)
Gary Tang is a research scientist at Netflix focused on driving long term member satisfaction through Netflix's core recommendation algorithms. His primary research interests are in bandits, reinforcement learning and off-policy evaluation. Gary holds a PhD from Stanford University.
Tonia Danylenko (Spotify)
Tonia Danylenko is a Senior Machine Learning Engineering Manager at Spotify, where she leads cross-functional teams working on AI-driven projects, including AI DJ, AI Playlists, and watch feeds, leveraging large language models (LLMs) to enhance these experiences as part of Spotify’s personalization mission. She previously led the development of a strategic and business-focused promotional platform at Spotify. Tonia is also a co-organizer of WiDS AI and ML in Sweden and holds a PhD in Computer Science. Her research interests are in theoretical computer science and AI in personalization and advertising.
Roberto Sanchis Ojeda (Spotify)
Roberto is the Director of Machine Learning at Spotify, leading teams that explore the boundary between ML/AI content recommendation products and business strategy. He focuses on empowering ML managers and Staff Engineers to deliver cutting edge tech solutions that support the need to optimize our recommendations towards long-term rewards and to explore and promote new content in a timely manner. Prior to Spotify, He worked as Data Scientists / ML researcher at Stitch Fix and Netflix, and he holds a PhD from MIT in Astrophysics.
Program
Session 1: 9:00-10:30
9:00-9:10 Welcome
9:10-9:45 Keynote-1: Optimizing for a broader world of entertainment (Gary Tang, Netflix)
9:45-10:00 Ranking Policy Learning via Marketplace Expected Value Estimation From Observational Data, Ehsan Ebrahimzadeh, Nikhil Monga, Hang Gao, Alex Cozzi and Abraham Bagherjeiran (eBay).
10:00-10:15 Beauty Beyond Words: Explainable Beauty Product Recommendations Using Ingredient-Based Product Attributes Celine Liu, Rahul Suresh and Amin Banitalebi-Dehkordi (Amazon).
10:15-10:25 Welfare-Optimized Recommender Systems Benjamin Heymann, Flavian Vasile and David Rohde (Criteo).
Coffee Break 10:30-11:15
Session 2: 11:15-12:45
11:15-11:50 Keynote-2: “Strategy meets Relevance: The Art of Combining Strategy and Relevance for Content at Spotify” Tonia Danylenko & Roberto Sanchis Ojeda (Spotify)
11:50-12:05 Combining Bundle Economy and Relevancy through Content Recommendations in Candy Crush Saga Styliani Katsarou, Francesca Carminati, Martin Dlask, Marta Braojos, Lavena Patra, Richard Perkins, Carlos Garcia Ling and Maria Paskevich. (King Sweden)
12:05-12:20 Learning Set Embeddings for Fashion Compatibility Recommendation, Indra Firmansyah, Randolf Scholz, Adrian Nahmendorff, Ngoc Son Le, Shereen Elsayed and Lars Schmidt-Thieme (University of Hildesheim).
12:20-12:30 A Hybrid Meta-Learning and MAB Approach for Context-Specific Multi-Objective Recommendation Optimization, Tiago Cunha and Andrea Marchini (Expedia Group).
12:30-12:40 Optimising Contextual Advertising through Real-Time Bidding with Budget Constraints, Jingwen Cai and Johanna Björklund (Umeå University).
12:40-12:45 Closing remarks
Google Sites
Report abuse