Mend the Measurement Gap: Latent User Preference Modeling for Short-Form Video Recommendation
A latent preference model lifts YouTube Shorts viewer enjoyment 2.67% by correcting watch-time confounders.
The Factorized Latent Value Model treats short-form video behaviors as noisy measurements of a low-dimensional latent preference, separating genuine value from confounders such as video duration and inattention. A restricted baseline captures predictable measurement artifacts, while a routed latent path estimates preference-relevant advantage for ranking. On YouTube Shorts, the model improved offline metrics and lifted a primary viewer enjoyment metric by 2.67% in online A/B tests.
- Models behaviors as noisy measurements of a latent preference state
- A baseline path absorbs duration, propensity, and session confounders
- Online A/B tests raised a primary enjoyment metric by 2.67%
- Latent score can serve as a ranking feature or score
- Products
- YouTube Shorts
- Organizations
- YouTube
- AI models
- FLVM
From the original article
1 min read at arxiv.orgRecommender systems rely heavily on heterogeneous behavioral feedback to infer user preference. Although abundant, these signals are imperfect measurements: the same observed behavior can arise from different underlying states, such as genuine enjoyment, passive consumption, or inattention.
By Shuo Chang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.