Published June 7, 2025
| Version v1
Software
Open
Artifact of the paper: Addressing Sensitivity Distinction in Local Differential Privacy : A General Utility-Optimized Framework
Description
This project is divided into two folders: main experimental project and frequent item mining
----------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------
# Main Experimental Project
# Project Structure
## Folders:
* **data**: Contains the real and synthetic datasets used in experiments, including raw datasets and intermediate files generated during data cleaning.
* **src**: Contains all experiment code.
* **src/protocols**: Implements three existing ULDP mechanisms (`uRR`, `uRAP`, `uHR`) and three new mechanisms proposed in this work (`uSS`, `uOUE`, `uOLH`).
## Python Files:
### Miscellaneous:
* `data_preprocessing.py`: Loads datasets into memory.
* `tools_function.py`: Provides two functions to compute `MSE` and `MAE`.
### Theoretical MSE Calculation:
* `theoretical_mse.py`: Computes the theoretical MSE of each mechanism according to the paper’s formulas.
* `withWithout_z.py`: Computes the theoretical MSE when `z=0`.
### ULDP Mechanisms (in `src/protocols`):
* `old_HR.py`: Existing ULDP mechanism `uHR`.
* `old_URAP.py`: Existing ULDP mechanism `uRAP`.
* `old_uRR.py`: Existing ULDP mechanism `uRR`.
* `uOLH.py`: Proposed mechanism `uOLH`.
* `uOUE.py`: Proposed mechanism `uOUE`.
* `uSS.py`: Proposed mechanism `uSS`.
### Data Cleaning:
* `Get_census_dataset.py`: Code to generate the Census dataset.
* `Get_Foursquare_dataset.py`: Code to generate the Foursquare dataset.
* `Get_normal_dataset.py`: Code to generate the Normal dataset.
### Main Experiment Scripts:
* `exp1_impact_of_epsilon.py`: Experiment 1 – Impact of varying `epsilon` on mechanism utility.
* `exp1_impact_of_sensitive.py`: Experiment 1 – Impact of varying sensitive data ratio on mechanism utility.
* `exp2_collaborative_sampling.py`: Experiment 2 – Impact of collaborative sampling on results.
* `exp3_without_z.py`: Experiment 3 – Impact of setting `z=0` on results.
* `exp4_mae.py`: Experiment 4 – Results when using `MAE` as the utility metric.
### Experiment Visualization:
* `experiment_visualization.py`: Plots experimental results.
1. Manually fill in MSE/MAE values from other experiments.
2. Ensure the output directory exists before running.
## Usage:
Run any main experiment script with:
```bash
python <experiment_script>.py
```
----------------------------------------------------------------------------------------------
----------------------------------------------------------------------------------------------
# Frequent Item Mining
# Project Structure:
## Folders:
* census-data: Contains the real dataset used for the frequent item mining task, including the full data file `census.txt` and the sensitive subset file `census_sen.txt`.
* foursquare-data: Contains the real dataset used for the frequent item mining task, including the full data file `foursquare.txt` and the sensitive subset file `foursquare_sen.txt`.
* results: Stores the data outputs from experiments.
* ULDP: Contains six perturbation mechanism implementations for evaluation, namely `old_HR.py`, `old_URAP.py`, `old_uRR.py`, `uOLH.py`, `uOUE.py`, and `uSS.py`. These scripts are called by the main experiment drivers.
## Scripts:
Main experiment drivers (`exp-<mechanism>-<dataset>.py`):
* `exp-uhr-cens.py`: Runs the UHR mechanism on the Census dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-uhr-fours.py`: Runs the UHR mechanism on the Foursquare dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-uolh-cens.py`: Runs the UOLH mechanism on the Census dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-uolh-fours.py`: Runs the UOLH mechanism on the Foursquare dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-uoue-cens.py`: Runs the UOUE mechanism on the Census dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-uoue-fours.py`: Runs the UOUE mechanism on the Foursquare dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-urap-cens.py`: Runs the URAP mechanism on the Census dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-urap-fours.py`: Runs the URAP mechanism on the Foursquare dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-urr-cens.py`: Runs the URR mechanism on the Census dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-urr-fours.py`: Runs the URR mechanism on the Foursquare dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-uss-cens.py`: Runs the USS mechanism on the Census dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
* `exp-uss-fours.py`: Runs the USS mechanism on the Foursquare dataset for frequent item mining and accuracy evaluation; saves results to the `results` folder.
## Utility modules:
* `data.py`: Loads data files and preprocesses datasets. Called by the main experiment scripts.
* `F1score.py`: Implements the F1 score evaluation metric. Called by the main experiment scripts.
* `NDCG.py`: Implements the NDCG evaluation metric. Called by the main experiment scripts.
* `ULDPFIM.py`: Implements the frequent item mining task: generates candidate itemsets from the full dataset and mines frequent items from those candidates. Called by the main experiment scripts.
## Environment:
* See environment.yml
* Run experiments by executing:
```bash
python3 exp-<mechanism>-<dataset>.py
```
For example:
```bash
python3 exp-uhr-cens.py
```
Files
Files
(1.5 GB)
| Name | Size | Download all |
|---|---|---|
|
md5:a908ea06cc29a1c9a09bc91ab86b45ea
|
1.5 GB | Download |