Protecting De-identified Documents from Search-based Linkage Attacks
The method counters search-based linkage attacks by finding short word sequences (N-grams) that appear in very few documents. It then uses an AI model to rewrite these problematic phrases repeatedly until they no longer link back to the original source. This effectively stops simple searches from connecting de-identified text to its origin while keeping the overall meaning intact.