Evaluating the Privacy Implications of Frequent Itemset Disclosure

Area of Science:

• Data Science
• Privacy-Preserving Data Mining

Background:

• Frequent itemset mining is crucial for data analytics, but releasing only itemsets raises privacy concerns.
• Evaluating the privacy implications of shared frequent itemsets is challenging.

Purpose of the Study:

• To define the k-distant-IFM-solutions problem for quantifying privacy risks.
• To find k transaction datasets with maximized pair distance to assess privacy.

Main Methods:

• Defined the NP-hard k-distant-IFM-solutions problem.
• Proposed a 2-approximate algorithm and faster heuristics.
• Evaluated solutions on real-world datasets.

Main Results:

• The proposed methods provide a way to evaluate privacy risks associated with frequent itemset disclosure.
• The degree of difference between reconstructed datasets serves as a privacy risk metric.

Conclusions:

• The k-distant-IFM-solutions problem offers a novel approach to privacy evaluation in frequent itemset mining.
• The developed algorithms and heuristics are effective in assessing privacy risks on real data.