Dynamic Prior Thompson Sampling for Cold-Start Exploration in Recommender Systems Conference: August, 2025; Canada;

Zhenyu Zhao, David Zhang, Ellie Zhao, Ehsan Saberian · 2026

Read Dynamic Prior Thompson Sampling for Cold-Start Exploration in Recommender Systems Conference: August, 2025; Canada; by Zhenyu Zhao, David Zhang, Ellie Zhao, Ehsan Saberian (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-00943.txt.

Description

Zhenyu Zhao and colleagues propose Dynamic Prior Thompson Sampling, controlling the chance that a new arm beats the incumbent winner. Cold-start exploration forces new or sparse items to gain traffic, but over-exploration harms users. Standard Thompson Sampling starts with a uniform Beta(1,1) prior, implying a 50% success rate for unseen items. Batched updates amplify the problem: for hours, new items stay no-data, so the prior dominates before feedback arrives. Their closed-form quadratic solution sets the prior mean enforcing P(Xj > Yk) = epsilon at introduction, making exploration tunable.

Catalog record

CREATORZhenyu Zhao, David Zhang, Ellie Zhao, Ehsan Saberian
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE32229
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00943

Browse the full library · About the Tower of Babel