Network A/B Testing with Non-Cooperative Behavior

This repo contains the code for our work on A/B testing in networks with non-cooperative behavior.

Cite our paper at USENIX Security 2022:

@inproceedings{clary2022noncooperative, 
    author={Clary, Kaleigh and Tosch, Emma and Onaolapo, Jeremiah and Jensen, David D.}, 
    title={Stick it to {T}he {M}an: Correcting for Non-Cooperative Behavior of Subjects in Experiments on Social Networks},
    journal={{USENIX} Security},
    year={2022}
}

Run the Experiments

REPLICATION.md contains installation, configuration, and run instructions for experiments and figures in the paper.

Overview

Given a network graph and simulation parameters identified in a configuration file,

• Cluster-randomized treatment assignment
• Non-cooperative participant network set construction
• Outcome simulation per n adversaries up to dominating set
• Model fit and outcome estimation with linear estimator
• Bias calculation in sample

Experiment Configs

Options:

• Network structure for A/B test
• Non-cooperative participant network placement (random RNCP, greedy GNCP)
• Outcome simulation parameters
• Non-cooperative outcome models

More information is provided in this README.

Network Graph Resources

We provide some existing graph structures for the experiments:

• Synthetic Corpus: We provide a corpus of graph types used for the simulation studies in the paper, generated under the same set of parameters.
• Real-World: We use networks released in the SNAP Library.
• Custom: Instructions for adding custom graph source.

Alternatively, run the scripts used to generate graphs in the corpus to generate new graphs. More details at this README.