Practical Multi-Task Learning for Rare Conversions in Ad Tech Conference: Proceedings of the Nineteenth ACM Conference on Recommender Systems; September 22–26, 2025; Prague, Czech Republic Proceedings of the Nineteenth ACM Conference on Recommender Systems (RecSys ’25), September 22–26, 2025, Prague

Yuval Dishi, Ophir Friedler, Yonatan Karni, Natalia Silberstein, Yulia Stolin · 2025

Read Practical Multi-Task Learning for Rare Conversions in Ad Tech Conference: Proceedings of the Nineteenth ACM Conference on Recommender Systems; September 22–26, 2025; Prague, Czech Republic Proceedings of the Nineteenth ACM Conference on Recommender Systems (RecSys ’25), September 22–26, 2025, Prague by Yuval Dishi, Ophir Friedler, Yonatan Karni, Natalia Silberstein, Yulia Stolin (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-2507-20161.txt.

Description

Yuval Dishi and co-authors share a deployed method that lifts ad forecasts for rare buy events, in a paper from 2025. Buys that fire on fewer than one in a hundred views starve a lone model of signal. The fix learns a shared base from all signals at once, then splits into twin towers, one for rare buys and one for common ones.

Live and past tests both back the split. Past data show a small but steady lift in rank score. In live use the cost per buy fell by about two in a hundred, a real save at ad scale. The authors stress that the build runs in full use, not just in lab tests, and holds its gain day to day.

Catalog record

CREATORYuval Dishi, Ophir Friedler, Yonatan Karni, Natalia Silberstein, Yulia Stolin
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE17 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-20161

Browse the full library · About the Tower of Babel