Your Recommender Can’t See New Items: Fixing Cold Start at the Root

​Most recommender systems perform until they encounter something new. Cold start is often treated as a data sparsity issue, but in practice it is a representation problem. Systems that rely on interactions alone are inherently blind to large parts of the catalog.In this workshop, we break down why common solutions fail and present a structured path toward more robust architectures. We cover approaches ranging from basic content features to dense representations and semantic identifiers, and show how they interact with state of the art recommendation models.Using concrete examples and a guided walkthrough, we demonstrate how Albatross incorporates richer item representations into recommendation systems to improve performance on new and long-tail items. Participants will leave with a clear mental model of the problem and a practical toolkit for addressing it in production environments.

​Please only apply for this workshop if you hold a valid GenAI Zürich 2026 ticket. Applications submitted without a valid ticket will be rejected. Don’t have a ticket yet? You can secure yours here****.

​All applications are subject to approval by the workshop organizers. If your application is successful, we recommend arriving 10 minutes before the workshop starts to secure your seat. Allocated seats that have not been filled 5 minutes before the workshop start may be released and reallocated to spontaneous participants.