Language Model Representations for Efficient Few-Shot Tabular Classification Conference: Proceedings of the ACM Web Conference 2026; April 13–17, 2026; Dubai, United Arab Emirates Proceedings of the ACM Web Conference 2026 (WWW ’26), April 13–17, 2026, Dubai, United Arab Emirates DOI: 10.1145/377490
Inwon Kang, Parikshit Ram, Yi Zhou, Horst Samulowitz, Oshani Seneviratne · 2026
Read Language Model Representations for Efficient Few-Shot Tabular Classification Conference: Proceedings of the ACM Web Conference 2026; April 13–17, 2026; Dubai, United Arab Emirates Proceedings of the ACM Web Conference 2026 (WWW ’26), April 13–17, 2026, Dubai, United Arab Emirates DOI: 10.1145/377490 by Inwon Kang, Parikshit Ram, Yi Zhou, Horst Samulowitz, Oshani Seneviratne (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-2602-15844.txt.
Description
Kang and colleagues ask whether already-deployed large language models can sort ordered data in web-native tables, skipping focused models or extensive retraining. The web holds rich ordered data in tables, from product catalogs and knowledge bases to scientific data sets. Their light method, Table Representation with Language Model, or TaRL, supports few-shot table sorting directly using meaning vectors of individual table rows.
Naive use of those vectors underperforms focused table models. But their potential unlocks with two key techniques: removing the common component from all vectors and tuning the softmax temperature. A simple learner, trained on hand-made features, learns to predict a right temperature. That approach achieves performance matching top models in low-data cases, with k at most 32, on tables rich in meaning. The findings show the worth of reusing existing language model setup for fast meaning-driven web table understanding.
Catalog record
| CREATOR | Inwon Kang, Parikshit Ram, Yi Zhou, Horst Samulowitz, Oshani Seneviratne |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-15844