// the take
QRec is an academic benchmark library that reimplements ~35 published recommender-system papers (BPR, NGCF, LightGCN, SimGCL, several GAN- and GCN-based social recommenders) behind a common config-file interface, built on TensorFlow 1.14. It's for researchers who need a fast way to reproduce or compare against published baselines, not for anyone building a production recommender.
The breadth is the real value: you get working, citable reimplementations of papers spanning SIGIR/KDD/WWW/AAAI from 2008 through 2022 in one place, with matching dataset formats (Ciao, Epinions, Douban, Yelp, Amazon-Book) already wired up. The base-class design (recommender.py, socialRecommender.py, graphRecommender.py, iterativeRecommender.py) means adding a new model mostly means overriding a handful of lifecycle methods, not fighting the framework. Config-driven experiment setup (conf files for train/test split, cross-validation, cold-start evaluation, top-N ranking) gives you reasonably rigorous evaluation protocols for free instead of hand-rolling them.
It's pinned to TensorFlow 1.14 and Python 3.7 — both long past end-of-life, and getting this running in a modern environment means fighting pip over ancient wheels or isolating it in a dedicated conda env. The 'what's new' log stops at SimGCL in March 2022, and the author already forked effort into a separate PyTorch library (SELFRec), which reads as this repo being in maintenance mode at best despite the Dec 2023 push. There's no packaging (no pip install, no setup.py mentioned) — you clone the repo and run main.py against a conf file, which is fine for a one-off experiment but painful if you want to pull individual models into another codebase. No test suite is mentioned anywhere, so correctness of any given algorithm rests entirely on trusting the implementation against the paper.
We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →