DynamiX: Dynamic Resource eXploration for Personalized Ad-Recommendations

Sohini Roychowdhury, Adam Holeman, Mohammad Amin, Feng Wei, Srihari Reddy · 2025

Read DynamiX: Dynamic Resource eXploration for Personalized Ad-Recommendations by Sohini Roychowdhury, Adam Holeman, Mohammad Amin, Feng Wei, Srihari Reddy (2025) 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-2511-18331.txt.

Description

Sohini Roychowdhury, Adam Holeman, Mohammad Amin, Feng Wei, Bhaskar Mehta, and Srihari Reddy present DynamiX, a framework that trims how ad recommendation systems chew through user histories. Full engagement histories are heavy to process and full of noise. DynamiX sorts user activity by session and by surface, using links between dwell time and conversion events to decide which recorded events carry real signal for each user segment.

Unhelpful events can then be dropped and useful features boosted, guided by self-supervised features built from the events themselves. The authors report modest gains in training and serving speed from removal, and a small quality gain plus faster serving from boosting, all without losing prediction accuracy. At ad scale, even small percentage gains translate into large savings.

Catalog record

CREATORSohini Roychowdhury, Adam Holeman, Mohammad Amin, Feng Wei, Srihari Reddy
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-18331

Browse the full library · About the Tower of Babel