One Word is Enough: Minimal Adversarial Perturbations for Neural Text Ranking

Tanmay Karmakar, Sourav Saha, Debapriyo Majumdar, Surjyanee Halder · 2026

Read One Word is Enough: Minimal Adversarial Perturbations for Neural Text Ranking by Tanmay Karmakar, Sourav Saha, Debapriyo Majumdar, Surjyanee Halder (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-2601-20283.txt.

Description

Tanmay Karmakar and three colleagues show that a single well-chosen word can fool a neural text ranking model. Ranking models are effective at retrieval, but earlier work showed they can be pushed around by attacks. The authors revisit this with a minimal, query-aware attack that promotes a target document.

The attack inserts or swaps in one word aligned with the query, the query center. If one word can lift a document's rank, ranking systems are more fragile than their scores suggest. The finding matters for search engines and for any contest where placement brings money or attention.

Catalog record

CREATORTanmay Karmakar, Sourav Saha, Debapriyo Majumdar, Surjyanee Halder
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE18 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-20283

Browse the full library · About the Tower of Babel