I used my distributed stochastic gradient descent matrix factorization to create a recommender for the 3.5M open source projects on GitHub. There were approximately 4.5M users who had interacted with these projects.
The utility/ratings matrix was created from