RecBole

Why we included this project

Researchers and engineers who build recommender systems will find RecBole hard to beat as a starting point. It ships with 94 PyTorch implementations of common recommendation algorithms, from general collaborative filtering to sequential, context-aware, and knowledge-based models, plus preprocessed copies of 44 benchmark datasets. Because everything shares the same data format and training pipeline, you can swap models and compare baselines without rewriting code or reconciling incompatible dataset schemas. That same pipeline handles data processing scripts, so raw logs turn into reproducible experiments quickly. For teams following newer research, the companion RecBole2.0 toolkits extend the library into directions like debiased and fairness-aware models.