USRW 2026 — Unified Search and Recommendation Workshop

About the Workshop

Search (IR) and recommendation (RecSys) have long been developed in parallel, despite serving the same discovery goal over shared item universes. The rise of Large Language Models and agentic systems is collapsing this divide, with unified architectures and generative retrieval increasingly powering both query-driven and feed-driven experiences.

The Unified Search and Recommendation Workshop (USRW) brings together researchers and practitioners from both the IR and RecSys communities to accelerate this unification. Our goals are to:

Define shared tasks and datasets over a common item universe
Compare unified architectures (e.g., bi-encoders vs. seq2seq/generative models with semantic IDs)
Align evaluation across intents, latency, cost, and safety
Foster cross-pollination between industry practitioners and academic researchers

This is the inaugural edition of USRW — a full-day workshop designed to establish unified search-and-recommendation as a mainstream research and production reality.

Keynote Speakers

Mounia Lalmas

Senior Director of Research @ Spotify

Building AI-Driven Experiences at Scale: When Search and Recommendation Become One Conversation

Abstract

Search and recommendation have historically been developed as separate systems, one driven by explicit queries, the other by implicit behavioural signals. Yet they serve the same fundamental goal: connecting people to content. Advances in generative models and agentic architectures are collapsing this divide, enabling unified systems that reason over intent rather than simply matching queries or predicting preferences.

In this keynote, I discuss how this convergence is beginning to play out in practice at Spotify, a platform serving over 760 million users across music, podcasts, and audiobooks. Drawing on recent work, I describe early steps toward making search and recommendation operate as a single, intent-aware experience, through generative models, steerable recommendations, and agentic architectures.
I reflect on what unification requires in practice, and on the open problems that remain as we move from systems that predict preferences to systems that understand and respond to what users actually want.

Bio

Mounia Lalmas is Senior Director of Research at Spotify, where she leads Tech Research in Personalisation. Her work develops algorithmic and methodological solutions that help users connect with content they will enjoy across music, podcasts, and audiobooks. Her research spans recommender systems, information retrieval, evaluation, and user modelling, with a current focus on applying generative models to large-scale online systems. She previously held senior roles at Yahoo, where she focused on advertising quality, user engagement, and search, and was Professor of Information Retrieval at Queen Mary University of London. Mounia is an Honorary Professor at University College London and a Distinguished Research Fellow at the University of Amsterdam. She has co-chaired SIGIR 2015, WWW 2018, WSDM 2020, and CIKM 2023, and is a member of the SIGIR Academy Class of 2024. She has authored over 270 publications and co-authored Measuring User Engagement (Morgan & Claypool).

Harrie Oosterhuis

Associate Professor @ University of Amsterdam

Learning to Rank Everything, Everywhere and All at Once - A Perspective on the Future of Search and Recommender System Optimization

Bio

Dr. Harrie Oosterhuis is an Associate Professor at the Informatics Institute of the University of Amsterdam. His research lies on the intersection of Machine Learning, Information Retrieval and Recommendation Systems and primarily concerns statistical methods for optimizing ranking models. Much of his work combines theory and practice; in recent years, Harrie has combined his academic position with part-time research positions at Google DeepMind, Twitter, and currently Amazon Music. He graduated his PhD cum laude at the University of Amsterdam in 2020 on his thesis titled "Learning from User Interactions with Rankings: A Unification of the Field" and since then he has received the WSDM‘21, SIGIR‘21 and ICTIR‘22 best paper awards, in addition to a 2021 Google Research Scholar Award, a 2023 Radboud Science Award, a 2024 Andreas Bonn medal, a 2024 ACM SIGIR Early Career Researcher Award and a 2022 NWO Veni Grant.

In-workshop Tutorial

Dive Deep: Unified Search and Recommendation

Why search and recommendation are rapidly converging, and why they are still not simply “the same task.”

Aleksandr V. PetrovResearch Scientist @ Spotify

Gustavo PenhaSenior Research Scientist @ Spotify

We will cover

Shared roots: the historical intersection of information retrieval, information filtering, and content-based recommendation.
Divergence: where search and recommendation split in signals, interfaces, evaluation, and serving.
Layers of unification: catalogues, logs, representations, candidate generation, ranking, interfaces, and evaluation.
Modern paradigms: joint models, semantic IDs, generative retrieval, language interfaces, and agentic orchestration.
Real-world trade-offs: negative transfer, modular vs. unified architectures, latency, cost, controllability, and evaluation.

Our goal is to give attendees a clear conceptual map for thinking about unified discovery systems — not as a slogan, but as a layered design space.

Organizers

Advisory Board

Contact

For questions, reach out at loading…