Instagram Network Analysis

This project explores the structure of my Instagram mutual-follow network using tools from web scraping, network science, and machine learning. I scraped mutual connections, built a social graph, and used graph neural networks (GNNs) to model social patterns and predict missing links.

Overview

• Data Collection: Using Playwright, I scraped all users I both follow and am followed by. Then, I gathered each mutual’s following list and filtered it to users already in the mutual set.
• Graph Construction: Built a directed graph from this data using NetworkX and PyVis for visualization.
• Network Analysis: Computed global and local metrics including centrality, clustering coefficient, and modularity (via Louvain community detection).
• Graph Machine Learning: Used PyTorch Geometric to train simple GNNs on the graph to predict connections and explore embeddings.
• Visualization: Created an interactive PyVis network that highlights structural features and community groupings.

Reproducible files (minimal set)

The repository currently contains the minimal files you need to reproduce the final analysis in these exact locations:

• playwright_saveLogin.py — Run this first to create a saved session (auth.json). Location: /playwright_saveLogin.py (repo root).
• playwright_getMutuals.py — Main data collection script: reads a CSV of mutuals and writes a graph JSON. Location: /playwright_getMutuals.py (repo root).
• csvs/mutuals.csv — Input CSV listing the mutual accounts to inspect. Location: /csvs/mutuals.csv.
• graphs/mutuals_graph.json — Output graph JSON produced by playwright_getMutuals.py. Location: /graphs/mutuals_graph.json.
• analysis.ipynb — Notebook that loads the graph JSON, runs analysis, and writes the interactive HTML visualization (this used to be graph_vis3.ipynb). Location: /analysis.ipynb (repo root).
• requirements.txt — Python dependencies used to reproduce the environment. Location: /requirements.txt (repo root).
• auth.json — Saved session/login file created by playwright_saveLogin.py. If present, location: /auth.json (repo root). Treat it as a secret; add it to .gitignore or move it to secrets/ if you plan to commit the repo.

If you've archived exploratory files into archive/, that's fine — just make sure the scripts above remain at these locations or update analysis.ipynb to point at the new paths.

Quick run-order (zsh)

Run these from the repo root (paths are exact as listed above):

# 1) create and activate a venv and install requirements
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# 2) save login/session (creates auth.json)
python3 playwright_saveLogin.py

# 3) create graph JSON from mutuals CSV
python3 playwright_getMutuals.py csvs/mutuals.csv --output graphs/mutuals_graph.json

# 4) open `analysis.ipynb` and run the cells that load `graphs/mutuals_graph.json`

Notes

• If analysis.ipynb includes hard-coded relative paths, open it and update them to the exact paths above.
• Add auth.json, csvs/large_scrapes/, and archive/ to .gitignore if you don't want to track them in git.

Technical Highlights

• GNNs implemented with PyTorch Geometric
• Louvain clustering and centrality measures
• Fully automated web scraping pipeline with Playwright
• Interactive network rendered with PyVis
• Clean separation of scraping, analysis, and modeling logic in Jupyter notebooks

Notebook

All analysis can be found in graph_vis3.ipynb

Blog Post

A full write-up of this project, with visuals and explanations, is available here:
Blog: Instagram Network Analysis

Inspiration

This project was originally inspired by this Medium post, but the codebase is entirely my own.

Disclaimer

This project was done for educational purposes. Please respect Instagram’s terms of service.