PinFM: Foundation Model for User Activity Sequences at a Billion-scale Visual Discovery Platform DOI: XXXXXXX.XXXXXXX Conference: XXX; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Information systems Recommender systems
Xiangyi Chen, Kousik Rajesh, Matthew Lawhon, Zelun Wang, Hanyu Li, Haomiao Li, Saurabh Vishwas Joshi, Pong Eksombatchai, Jaewon Yang, Yi-Ping Hsu, Jiajing Xu, Charles Rosenberg · 2025
Read PinFM: Foundation Model for User Activity Sequences at a Billion-scale Visual Discovery Platform DOI: XXXXXXX.XXXXXXX Conference: XXX; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Information systems Recommender systems by Xiangyi Chen, Kousik Rajesh, Matthew Lawhon, Zelun Wang, Hanyu Li, Haomiao Li, Saurabh Vishwas Joshi, Pong Eksombatchai, Jaewon Yang, Yi-Ping Hsu, Jiajing Xu, Charles Rosenberg (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-2507-12704.txt.
Description
Xiangyi Chen and a team of co-authors presented a foundation model for the streams of clicks and views that drive a giant visual discovery platform, in a 2025 paper. Such activity sequences are among the strongest signals a recommender can read. The model must score millions of items per second, stay cheap and fast, and handle brand-new items.
Their model, PinFM, pre-trains a transformer of over 20 billion parameters on platform activity, then fine-tunes it per app. A new attention design that dedupes shared work raised throughput sixfold on internal data. Shaping input sequences let the model learn how activity and candidate items interact, which lifted engagement with new items by 20 percent. The system now serves over half a billion users.
Catalog record
| CREATOR | Xiangyi Chen, Kousik Rajesh, Matthew Lawhon, Zelun Wang, Hanyu Li, Haomiao Li, Saurabh Vishwas Joshi, Pong Eksombatchai, Jaewon Yang, Yi-Ping Hsu, Jiajing Xu, Charles Rosenberg |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-12704