recommenders-team/recommenders

Recommenders is an open-source library of examples and best practices for building recommendation systems, the kind of software behind "people who liked this also liked..." widgets, personalised news feeds, and movie suggestions on streaming services. Rather than being a single algorithm, it is a curated toolkit, originally seeded by Microsoft and now maintained as a project under the Linux Foundation of AI and Data, that helps developers prototype, experiment with, and deploy recommenders. The bulk of the repository is Jupyter notebooks organised around five steps in the lifecycle of a recommender: preparing data, training models, evaluating accuracy with offline metrics, tuning hyperparameters, and operationalising them in production on Azure. "Quick start" notebooks show the smallest amount of code to try an algorithm and "deep dive" notebooks walk through the underlying math. Alongside the examples, the repository ships utility code for loading common public datasets, splitting them into training and test sets, and computing evaluation metrics, plus implementations of many algorithms, collaborative filtering methods like Alternating Least Squares and Bayesian Personalized Ranking, deep-learning approaches like xDeepFM and DKN, that work in CPU, GPU, or Spark environments. You would reach for it if you are starting a recommender project and want known-good baselines rather than reinventing each algorithm yourself. The core package is written in Python and installable from PyPI; extras add GPU, Spark, development, and experimental options. The README suggests using uv for environment management and Python 3.11. The full README is longer than what was provided.