Code for the paper Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media.

1. To get started, create a Conda environment: conda create -n twitter python=3.9.11
2. Then, activate the Conda environment: conda activate twitter
3. Finally, install all requirements: pip install -r requirements.txt

To request access to our dataset, you can use this form.

Descriptions of the data files can be found in DATA.md.

• effects.ipynb: Average treatment effect (ATE) graph (Figure 1) and political tweets by in-group and out-group (Figure 2)
• demographics.ipynb: User demographics and ANES data comparison (SM section S2)
• metadata_stats.ipynb: Descriptive statistics for tweet metadata and individual user-level amplification (SM sections S4.1 & S4.2)
• likert_distributions.ipynb: Distribution of responses to Likert survey questions (SM section S4.3)
• outcomes_by_rank.ipynb: Outcomes broken down by the position of tweet in each timeline (SM section S4.4)
• effects_by_tweet_threshold.ipynb: ATE calculations across varied tweet thresholds (SM section S4.5)
• gpt_judgements.ipynb: ATEs for tweet outcomes calculated with GPT-4 labels (SM section S4.6)
• effects_het.ipynb: Effects calculated when restricting to subpopulations of users (SM section S4.7)

If you use our code or data, please cite our paper:

@article{milli2023twitter,
  title={Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media},
  author={Milli, Smitha and Carroll, Micah and Wang, Yike and Pandey, Sashrika and Zhao, Sebastian and Dragan, Anca D.},
  journal={arXiv preprint arXiv:2305.16941},
  year={2023}
}