Explore the effects of takedown delay on the persistence of illegal content

This repository contains code to reproduce the results in the paper ``Delayed takedown of illegal content on social media makes moderation ineffective''.

The model is an extension of SimSoM: A Simulator of Social Media

Overview of the repo

1. data: contains raw & derived datasets
2. example: contains a minimal example to start using the SimSoM model
3. libs: contains the extended SimSoM model package that can be imported into scripts
4. report_figures: experiment results, supplementary data and .ipynb noteboooks to produce figures reported in the paper
5. workflow: scripts to run simulation and Snakemake rules to run sets of experiments

Install

• This code is written and tested with Python>=3.6
• We use conda, a package manager to manage the development environment. Please make sure you have conda or mamba installed on your machine

Using Make (recommended)

To set up the environment and install the model: run make from the project directory (SimSoM)

Using Conda

1. Create the environment with required packages: run conda env create -n simsom -f environment.yml to
2. Install the SimSoM module:
   • activate virtualenv: conda activate simsom
   • run pip install -e ./libs/

Data

The empirical network is created from the Replication Data for: Right and left, partisanship predicts vulnerability to misinformation, where:

• measures.tab contains user information, i.e., one's partisanship and misinformation score.
• anonymized-friends.json is the adjacency list.

We reconstruct the empirical network from the above 2 files, resulting in data/follower_network.gml. The steps are specified in the script to create empirical network

Running the code

Check out example to get started.

• Example of the simulation and results: example/run_simulation.ipynb

Reproduce results from the paper:

1. From the root directory, unzip the data file: unzip data/data.zip -d .
2. Create config files specifying parameters for simulations: workflow/scripts/make_config.py
   • See example/data/config.json for example of a config file
3. Run a Snakemake rule corresponding to the simulations of interest.
   • e.g.: workflow/rules/vary_tau.smk reproduces the main results by varying tau, the illegal content half-life parameter.

Notes

The results in the paper are based on averages across multiple simulation runs. To reproduce those results, we suggest running the simulations in parallel, for example on a cluster, since they will need a lot of memory and CPU time.