Truthful Text Sanitization Guided by Inference Attacks Journal: Information Sciences
Ildikó Pilán, Benet Manzanares-Salor, David Sánchez, Pierre Lison · 2024
Read Truthful Text Sanitization Guided by Inference Attacks Journal: Information Sciences by Ildikó Pilán, Benet Manzanares-Salor, David Sánchez, Pierre Lison (2024) 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-2412-12928.txt.
Description
Ildiko Pilan and co-authors present a method for scrubbing personal details from text while keeping the text truthful, in a 2024 paper. Scrubbing must strike a balance: hide private facts, yet keep as much real content as it can. Their approach swaps private spans for broader terms that still hold true. A tuned large language model drafts swaps at several levels of breadth, then ranks them.
Each candidate swap is then attacked by the same model, which tries to guess back the private fact. Only swaps that resist the attack and stay informative are kept. The team also offers scores for privacy and utility that need no hand-tagged data. On a shared test set, their method kept more utility than full deletion, with a rise in re-identification risk of under one point. It also stayed more truthful than a well-known tool that invents fake replacements.
Catalog record
| CREATOR | Ildikó Pilán, Benet Manzanares-Salor, David Sánchez, Pierre Lison |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 66 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-12928