GitHub - recommenders-team/recommenders: Best Practices on Recommendation Systems

Abstraction: Microsoft open-source repository of recommendation algorithm best practices

Key points:

• 20,000+ GitHub stars; under Linux Foundation of AI and Data; covers 5 tasks: data prep, modeling, evaluation, hyperparameter optimization, operationalization on Azure
• 30+ algorithms including ALS, NCF, SAR, NRMS, LSTUR, NAML, NPA, DKN, xDeepFM, LightGCN, BPR, BiVAE, SASRec, Wide and Deep, VAE variants — many invented/contributed by Microsoft (marked *)
• Benchmark on MovieLens 100k (15 epochs): BiVAE MAP=0.146 best; BPR MAP=0.133; NCF MAP=0.108; ALS MAP=0.005 (underperforms at CF ranking)
• Recommended setup: uv for env management (10-100x faster than conda/pip), Python 3.11; GPU/Spark extras available
• Provides utilities for data loading, evaluation (MAP, nDCG@k, Precision@k, Recall@k, RMSE), train/test splits, and production deployment

Connections: Microsoft · Microsoft Recommenders · Github · Recommendation Systems · Collaborative Filtering · Content Based Filtering

Source: https://github.com/Microsoft/Recommenders