.png){kind=icon}
.png){kind=icon}
Return
Truthful text sanitization guided by inference attacks
.png){kind=icon}
OA
AI
DOI：10.1016/j.asoc.2025.114013.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose INTACT, a novel text sanitization approach using instruction-tuned LLMs. • INTACT replaces disclosive terms with truth-preserving generalizations. • INTACT’s re-identification risk is comparable to suppressing sensitive text. • INTACT outperforms related works in downstream utility preservation. • Manual evaluation shows INTACT excels in producing truthful generalizations.
Keywords:
Data privacy
Text sanitization
Data utility
Truth-preserving replacements
Large language models
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
6.6
Papers:
1.4W
Citations:
4.8W
