Social Bot Detection through Structural Holes and Centrality Measures: Analyzing the Position of Humans and Bots in Networks

Authors

Boen Liu – Brainstorming, Writing, Coding, Social Network Analysis, Visualizations

Disclaimer

This project was submitted for STATS201: Machine Learning for Social Science, instructed by Prof. Luyao Zhang at Duke Kunshan University in Spring 2025.

Acknowledgments

• Special thanks to Prof. Luyao Zhang for valuable suggestions and guidance throughout the project.
• Appreciation also goes to my two classmates for their insightful feedback and revisions.
• This project benefited from ChatGPT and Whimsical for language refinement and creating flowcharts.
• Thanks to open-source software such as numpy, pandas, easygraph, networkx, torch, and other libraries for coding and analysis.

Statement of Intellectual and Professional Growth

This project has been a valuable learning experience, teaching me how to systematically and rigorously conduct research—from formulating research questions to identifying appropriate datasets, generating ideas, conducting experiments, and writing up findings. The process also enhanced my skills in social network analysis, visualization, and applying machine learning to social science problems.

Poster

Table of Contents

Section	Description
code	Contains all scripts for explanation(main), prediction, and causal inference.
code/Explanation.ipynb	Script for Social Network Analysis.
code/causal_inference.ipynb	Script for Regression Discontinuity Design (RDD) analysis.
code/prediction.ipynb	Script for Bot Detection Model.
data	Stores the dataset used in this study.
visualizations	Includes all generated plots and visual representations.
docs	Contains the final report and related documentation.
docs/Final-Report.pdf	The final report for the study.

Code Execution

To run the code, follow these steps:

1. Clone the repository

   git clone https://github.com/Rising-Stars-by-Sunshine/Boen-Final-Project.git
   cd Boen-Final-Project

2. Install dependencies using pip:

   pip install -r requirements.txt

   or to install specific libraries individually, use:

   pip install networkx==2.5 Easygraph==1.4.1 pandas==1.3.2 statsmodels==0.13.2 matplotlib==3.4.3 seaborn==0.11.2 torch==1.9.0 numpy==1.21.1 sklearn==0.24.2

3. Navigate to the code directory

   cd code

4. Run the Jupyter Notebook

   • For explanation analysis: Open and run Explanation.ipynb
   • For prediction tasks: Open and run prediction.ipynb
   • For causal inference: Open and run causal_inference.ipynb

Dependencies

Required Libraries and Versions:

• Python 3.9
• networkx==2.5
• easygraph==1.4.1
• pandas==1.3.2
• statsmodels==0.13.2
• matplotlib==3.4.3
• seaborn==0.11.2
• torch==1.9.0
• numpy==1.21.1
• torch.nn (included in the torch library)
• sklearn==0.24.2

Example Usage

1. Running Social Network Analysis(Main):

   • For social network analysis, run the following notebook to analyze centrality measures and structure holes for bots and humans:

   jupyter notebook Explanation.ipynb

2. Training and Evaluating Predictive Models:

   • To train a hypergraph neural network model for bot detection, run the following:

   jupyter notebook prediction.ipynb

3. Running Causal Inference Analysis:

   • To perform Regression Discontinuity Design (RD) analysis on the impact of Twitter’s subscription program, execute:

   jupyter notebook causal_inference.ipynb