GitHub Repository Recommendations

A Chrome extension that shows similar repository recommendations directly in the GitHub sidebar.

Recommendations are based on how people star repos: if users who starred repo A also tend to star repo B, those repos are likely related.

Install the extension

1. Clone this repo:

   git clone https://github.com/ysz/gh-recommendations.git

2. Open chrome://extensions in Chrome
3. Enable Developer mode (top right)
4. Click Load unpacked and select the cloned folder
5. Open any repository on GitHub -- a "Similar Repositories" panel appears at the bottom of the sidebar

Star counts and descriptions are loaded live from GitHub.

Currently covers repositories with 100+ stars (~9K repos in the dataset).

How recommendations work

Three signals are combined to score each recommendation:

1. Co-star similarity (40%) -- collaborative filtering on star data. Two repos are similar if many of the same users starred both.
2. Description embeddings (35%) -- semantic similarity of repo descriptions using Qwen3-Embedding-0.6B (1024-dim vectors).
3. Topic overlap (15%) -- Jaccard similarity between GitHub topic tags.

Star data comes from GH Archive via Google BigQuery. Repo metadata comes from the GitHub REST API.

Project structure

manifest.json            # Chrome extension manifest
content.js               # Injects recommendations into GitHub sidebar
styles.css               # Minimal styling overrides
icons/                   # Extension icons
data/
  recommendations.json   # Pre-computed recommendations (~3 MB)
engine/                  # Recommendation pipeline (Python)
  main.py                # CLI: collect, build, recommend, evaluate
  collector.py           # Data collection (GitHub API + BigQuery)
  graph.py               # Sparse matrix construction
  recommender.py         # Hybrid recommendation engine
  evaluate.py            # Held-out evaluation (Hit Rate, MRR)

Rebuilding recommendations

The engine/ folder contains the Python pipeline that generates data/recommendations.json. You only need this if you want to rebuild the data yourself.

cd engine
uv sync        # or: pip install -e .

# Get star data from BigQuery (prints SQL to run in console.cloud.google.com)
python main.py bigquery-sql

# Load exported CSV
python main.py load-bigquery stars.csv

# Build graph + embeddings + export
python main.py build
python main.py embeddings
python main.py export-web

Requires Python 3.12+.

Data sources and attribution

• GH Archive -- public GitHub event data, available on Google BigQuery.
• GitHub REST API -- repository metadata. Subject to GitHub Terms of Service.
• Qwen3-Embedding-0.6B -- embedding model by Alibaba/Qwen (Apache 2.0).

License

MIT