Mining Frequent Patterns Through Microaggregation in Differential Privacy

Frequent pattern mining has been widely employed to analyze transaction datasets, but the question of how sensitive information contained in a dataset should be protected remains remains relatively unanswered. The differential privacy model provides a robust privacy guarantee, but the k-anonymity model provides better dataset utility. In this paper, a synergetic approach is proposed to simultaneously protect privacy and enhance data utility when mining top-k frequent patterns. First, microaggregated data is released, which achieves kanonymity, regardless of the query types the user may be using. Second, top-k frequent patterns are selected based on microaggregated data using the exponential mechanism. Finally, the true support of each top-k frequent pattern is perturbed by adding Laplace noise.