What it does

DeepRec collects implementations of neural recommendation models—AutoRec, NeuMF, CML, NFM, CASER, and others—into a single Python/TensorFlow framework. It covers three standard tasks: rating prediction, top-N item ranking, and sequential recommendation. The authors’ stated goal is saving researchers the grunt work of reproducing published baselines.

The interesting bit

The library is essentially a curated, runnable bibliography. Each model maps to a specific conference paper from the mid-2010s recsys boom, making it useful for comparing against “the paper said X” claims. The modular structure is meant to let you slot in new models without rewriting the evaluation plumbing.

Key highlights

• Bundles 10+ classic architectures (I/U-AutoRec, CDAE, NeuMF, CML, LRML, NFM, NNMF, PRME, CASER, AttRec)
• Three evaluation entry points: test_item_ranking.py, test_rating_pred.py, testSeqRec.py
• Published and citable—backed by a Recommender Systems Handbook chapter and an ACM CSUR survey
• GNU GPL licensed

Caveats

• Stuck on TensorFlow 1.7+; the README explicitly lists “Update to TensorFlow 2.0” as a todo, so migration work is on you
• Some models are marked draft status (LRML is noted “DRAFT version”)
• “Code refactoring” is also on the todo list, which suggests the internals may show their age

Verdict

Worth a look if you’re writing a recsys paper and need trustworthy baseline implementations of pre-2020 models. Skip it if you want modern TF/Keras, production-ready pipelines, or anything beyond academic reproducibility.