FLUID: From Ephemeral IDs to Multimodal Semantic Codes for Industrial-Scale Livestreaming Recommendation

TL;DR

FLUID introduces an ID-free, multimodal encoding framework for livestreaming recommendation systems, effectively addressing cold-start issues and improving online performance metrics at scale.

Contribution

It is the first to fully replace item IDs with hierarchical semantic codes in production-scale livestreaming recommenders, enhancing cold-start handling.

Findings

Achieved +0.55% Quality Watch Duration

Achieved +2.05% Cold-Start Room Views

Achieved +0.05% Active Hours

Abstract

Modern recommender systems rely heavily on ID-based collaborative filtering: each item is represented by a unique ID embedding that accumulates collaborative signals from user interactions. Livestreaming recommendation, however, faces a unique challenge in this paradigm: a live room typically broadcasts for only tens of minutes, so its item ID remains poorly learned in a persistent cold-start state and ID-centric ranking models fail to generalize. We present FLUID, the first framework to fully retire the candidate-side item ID from a production-scale livestreaming ranker. FLUID couples a cross-domain multimodal encoder, jointly trained on short videos and livestreams to produce discrete hierarchical codes (LUCID), with a late-fusion, ID-free design that injects slice-level and room-level LUCID as independent tokens, stabilized by a staged warmup under online incremental training. Deployed on our industrial livestreaming recommenders with a cross-platform combined user base of over one billion globally, FLUID delivers significant online gains of +0.55% Quality Watch Duration, +2.05% Cold-Start Room Views, and +0.05% Active Hours.