Eight deep-learning models and one very small bird

What it does

SparrowRecSys is a movie recommendation system built as a mixed-language Maven project. It ties together TensorFlow for model training, Spark for offline data processing, and a Jetty server for online serving, all wired to a browser frontend. The dataset is a trimmed MovieLens slice—1,000 movies—so the whole stack stays lightweight enough to run locally and step through.

The interesting bit

The name references the Chinese idiom “麻雀虽小，五脏俱全” (small sparrow, fully equipped), and the project takes it literally: it implements eight distinct deep-learning models—from Word2vec and DeepWalk to DIN and DeepFM—inside one coherent architecture rather than as isolated notebooks. That makes it unusual among teaching repos, because you can trace a paper from citation to production inference without switching repositories.

Key highlights

• Implements eight models including Wide&Deep, Two Towers, Neural CF, and DeepFM
• Follows a classic industrial architecture: offline batch processing, near-line stream handling, online model serving, and frontend display
• Ships with a curated MovieLens subset (1,000 movies) but supports the full 20M dataset
• Bundles an extensive reading list of foundational CTR and recommendation papers
• Tied to an associated course that walks the codebase line by line

Caveats

• The README and documentation are in Chinese, so English speakers will need translation help
• Requires a polyglot toolchain (Java 8, Scala 2.11, Python 3.6+, TensorFlow 2.0+), which is a lot of moving parts for a learning project
• Explicitly designed for learning and reference, not as a production framework

Verdict

Grab it if you are trying to bridge the gap between recommender-system papers and production plumbing. Skip it if you need a drop-in, scalable service or prefer a monolingual Python stack.