Enhances GitHub by showing similar projects in a repository's sidebar
- Provides similar repositories for all repositories with more than 100 stars
- Adds recommendations on your home page based on your latest stars
- Recommends new elements for a specific star list
Recommendations are generated by locating the nearest neighbors of a given repository within a vector space, where similar repositories are positioned close to each other. This vector space was built by training an SVC model on a large dataset containing over 300 million GitHub stars. To keep the model up-to-date, the dataset is refreshed incrementally — one-twelfth is updated each month.
Although recommendations were generated locally at first, some users experienced serious performance issues. Hence, I moved the recommendation engine to a server running Qdrant, a vector database. Qdrant implements lookup techniques more efficient than the brute-force I implemented initially. This allows getting fast results without impacting user's machines.
This extension is licensed under GPL v3. See CONTRIBUTING.md to get started!