What does t-closeness require, and what weakness of l-diversity does it fix?

Question

Accepted Answer

t-closeness requires the sensitive-attribute distribution within each group to stay within distance t of its distribution in the full dataset — fixing skewness and similarity leakage.
l-diversity ensures variety but ignores distribution, leaving skewness and similarity attacks open. t-closeness tightens the rule: the distribution of the sensitive attribute inside any equivalence class must closely mirror its distribution across the whole dataset (the gap bounded by a threshold t).
If every class looks statistically like the overall population, then learning which class someone is in tells you no more than the population baseline already did — neutralizing both the rare-value (skewness) and semantic-clustering (similarity) leaks.
The cost: forcing each group to match the global distribution usually demands heavy generalization, so t-closeness can hit utility hard.
Tip: The progression in one line — k-anonymity (group size) → l-diversity (variety of secrets) → t-closeness (the shape of the secrets matches the whole).
Go deeper:

t-closeness (Wikipedia) — distribution-matching within distance t of the global distribution.