finds.dev

wzhe06/SparrowRecSys

A Deep Learning Recommender System

View on GitHubHomepage ↗

★ 2,782PythonApache-2.0updated Jun 2024

// the take

SparrowRecSys is a teaching reference implementation of an industrial-style movie recommender, built as the companion code for a paid Chinese-language GeekTime course. It walks through the full pipeline end to end: Spark offline feature engineering, TensorFlow model training, a Redis-backed feature store, and a Jetty server for online serving. It's for engineers who want to see how the classic recsys papers (Wide&Deep, DeepFM, DIN, two-tower, etc.) map onto actual service code rather than just a notebook.

Implements a genuinely wide range of papers as runnable code - Word2vec/DeepWalk embeddings, Wide&Deep, NeuralCF, DeepFM, DIN, DIEN, two towers - each one linked directly to its source paper in the README, so you can read the paper and the implementation side by side. It covers the whole pipeline, not just model training: Spark for offline feature engineering, TensorFlow for the models, Redis for the feature store, and a Jetty server for online serving, which is the part most recsys tutorials skip. It ships with a trimmed MovieLens dataset and pretrained SavedModel checkpoints under webroot/modeldata, so you can run the online demo immediately without training anything first.

The stack is pinned to Java 8, Scala 2.11, and TensorFlow 2.0+, all effectively abandoned upstream; getting this running on a current JDK or TF version will take real work despite the mid-2024 last push. The detailed walkthrough lives behind a paid Chinese-language course, and the README itself is Chinese-only, so non-Chinese-speaking adopters are left to reverse-engineer intent from code and comments alone. There's no visible test suite, which is a real gap for something meant to teach production patterns - there's no way to verify a model or Spark pipeline change didn't break something. The online serving layer (Jetty plus flat CSV embedding files) is a toy approximation of production serving, not something you'd scale past the 1000-movie demo dataset.

We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →

// at a glance

Stars
2,782
Forks
865
Language
Python
License
Apache-2.0
Last commit
2 Jun 2024

// nearby

If SparrowRecSys isn't quite it.

All repos →