PrivBasis is a differentially private frequent itemset mining algorithm, which leverages a novel notion called basis sets. A theta-basisset has the property that any itemset with frequency higher than theta is a subset of some basis. PrivBasis first privately constructs a basis set and then uses it to find the most frequent itemsets.
For more details, please see our paper: Ninghui Li, Wahbeh H. Qardaji, Dong Su, Jianneng Cao: PrivBasis: Frequent Itemset Mining with Differential Privacy. PVLDB 5(11): 1340-1351 (2012).
- The dataset is assumed to be space separated.
- The max clique size (or basis length) is less than 12 as we claimed in the Section 4.2.
The entrance of the experiment is the PrivBasisMain.py file. Please go to line 300 to change the dataset path, minimum support count, repeat time and k values. When the experiment is done, the evaluation result will be printed in the terminal.
- Python 2.7.12
- Numpy 1.12.1
- Pandas 0.19.1
Maintainer: Dong Su, sudong.tom@gmail.com