Research Of Frequent Itemsets Mining That Supports Differential Privacy

Frequent itemsets mining from a transaction dataset has been well studied in both the database and data mining community for many years,one popular scenario is that if the dataset contains private information,publishing frequent itemsets may pose tremendous threats to individual's privacy.DPFP-Growth proposes a frequent itemsets mining approach under differential privacy which achieves high efficiency and high data utilities.DPFP-Growth proposes a new structure TFP-tree that is convenient to protect the private information of nodes.Before constructing TFP-tree,DPFP-Growth truncates transactions that is based on exponential mechanism.After constructing TFP-tree,DPFP-Growth transforms TFP-tree to a FP-tree.A constrained least squares problem is proposed to enforce the consistency constraints of noisy FP-tree.Then DPFP-Growth utilizes FP-Growth to mine frequent itemsets.Experiment results on real dataset reveal that DPFP-Growth can mine frequent itemsets with high utilities and high efficiency compared with other approaches.DPFP-Growth also compares DPFP-Growth method with adding laplace noise to FP-tree directly,the results show that DPFP-Growth has higher utilities than adding laplace noise to FP-tree directly.The experiment results of mining representative patterns show that mining representative patterns under differential privacy produces similar number of patterns with RPlocal and the efficiency of mining representative patterns under differential privacy algorithm is almost same as that of RPlocal.