WGT-LDP

This paper is published at VLDB 2025. This repository contains the code for the manuscript:

Continuous Publication of Weighted Graphs with Local Differential Privacy

Requirements

To run the experiments in this repository, you need numpy, networkx, scikit-learn. You can install all the packages is through conda and pip:

pip install numpy
pip install networkx
pip install scikit-learn

File Contents

• data is the folder with all datasets.
  • EmailDept1_LDP is the Email-Eu dataset in our paper, which contains dynamic communications between 319 nodes over 173 time steps.
  • Forum_LDP is the Forum dataset in our paper, which contains dynamic interaction records between 899 students over 24 time steps.
  • Tech_LDP is the Tech-AS dataset in our paper, which contains dynamic connections between 5000 autonomous systems over 24 time steps.
• result is the folder for storing synthetic weighted graph (.txt format), including two examples.
  • SynGraph_Save stores the average results of ten runs on the Forum dataset when the privacy budget is 1.
  • SynGraph_Save_vary_w stores the average results of ten runs on the Forum dataset when the sliding window size is 6.
• main.py is the file used to run the WGT-LDP framework with different privacy budgets.
• main_vary_w.py is the file used to run the WGT-LDP framework with different sliding windows.
• main_event.py is the file used to run the WGT-LDP framework with event-level privacy.
• utils.py includes some functions that are needed for other files.
• IM_spread_LDP_eps.py is used to obtain the results of influence maximization under different privacy budgets.
• IM_spread_LDP_win.py is used to obtain the results of influence maximization under different sliding windows.

Running

I. main Experiments

you can run main experiments with python {file name}.py.

###### Example 1 ######
python main.py

###### Example 2 ######
python main_vary_w.py

###### Example 3 ######
python main_event.py

Default dataset: Forum (pre-configured).

To switch datasets (e.g. to Tech-AS):

• Comment out Forum block in your target main_*.py file:

# dataset: Forum
# data_path = "./data/Forum_LDP/FbForum"
# node_num = 899
# snapshot_num = 24
# max_h = 168

• Uncomment target dataset block:

# dataset: Tech-AS
data_path = "./data/Tech_LDP/tech"
node_num = 5000
snapshot_num = 24
max_h = 24

II. Influence Maximization Experiments

you can run influence maximization experiments with python {file name}.py.

###### Example 1 ######
python IM_spread_LDP_eps.py

###### Example 2 ######
python IM_spread_LDP_win.py

Preparation (required before running IM experiments):

1. Run a main experiment with graph saving enabled:

• In the main experiment file (e.g., main.py), uncomment:
  • Target dataset_name assignment.
  • utils.save_graph_with_params(...) call.

###### Example ######
# Before:
# dataset_name = "tech"
# utils.save_graph_with_params(dataset_name, epsilon, ...)
# After:
dataset_name = "tech"
utils.save_graph_with_params(dataset_name, epsilon, ...)

Note: Run main experiment after this change to generate synthetic graph data (.txt format).

2. Configure the IM experiment file:

• In IM_spread_LDP_eps.py or IM_spread_LDP_win.py, set dataset_name to match:

# For Tech-AS:
dataset_name = 'tech'

Important Notes

• Dataset Consistency: The dataset name must be identical in the main experiment (when saving) and the IM experiment.
• File-Specific Configuration: Each main experiment file (main.py, main_vary_w.py, main_event.py) has its own configuration block. If you switch datasets, you must update the configuration in the specific main file you are running.
• Graph Saving: Only uncomment the saving code when you intend to generate data for IM experiments (it may slow down the main experiment).