RecSys-Industrial-Book

简体中文说明 · Browsable chapter and concept catalog · Book PDF reader · Chapter code

An open Chinese-language book, 推荐系统：工业架构与核心算法 (Industrial Architecture and Core Algorithms of Recommender Systems), and a local-first learning toolkit. It is intended for graduate students, recommender-system engineers, architects, and researchers with prior ML or recommendation experience. The book PDFs and most learning materials are in Chinese; there is no English translation of the book.

What you can explore

Industrial recommender-system architecture: data and features, candidate retrieval, pre-ranking, ranking, re-ranking, online serving, evaluation, and reliability.
Models and methods: collaborative filtering, two-tower retrieval, approximate nearest neighbors, deep ranking, multi-task learning, bias, duration prediction, and other topics in the published chapters.
Book-grounded learning: the PDF reader, static chapter and concept catalog, curated concept cards with PDF-page references, and learning paths. The released chapter PDFs are authoritative; navigation cards are not a substitute for the book.
Engineering examples: chapter source-code pages and runnable templates for practicing recommendation workflows. Teaching examples marked [MOCK] are not production integrations; the vector-version exercise ranks fixed vectors exactly, without ANN indexing or model training.
Local AI Agent toolkit: PDF-page-cited retrieval and Q&A, chapter-code discovery, study plans, self-evaluation, and engineering exercises. Search indexes and personal progress stay on your machine, not on the public site.

For specific engineering questions, start with the Chinese concept cards on two-tower retrieval, ANN index choice, feature freshness, or online fallback. Each links to the released PDF page and separates book-grounded claims from additional engineering checks; these are navigation aids, not an English translation.

Try it locally

With Python 3.10+ and pdftotext installed, run from the repository root:

python3 tools/recsys_agent.py index
python3 tools/recsys_agent.py doctor
python3 tools/recsys_agent.py search "two-tower negative sampling bias"
python3 tools/recsys_agent.py code "two-tower" --chapter 10

For commands and citation rules, see the Agent guide and repository instructions. The public website is a static reader, not a hosted AI assistant.

Sources, license, and linking

The released PDFs are the source for book claims. When discussing a topic, link to its concept card or chapter map and cite the corresponding PDF chapter and page; when discussing an implementation, link to the specific code page. Do not treat a code example as evidence for a book claim. You may link to this repository and its public pages; check the license before reusing or redistributing content (CC BY-NC-SA 4.0). This repository does not publish its private LaTeX source, bulk PDF text, or a prose-equivalent Markdown book.

Canonical project entry: https://jhljx.github.io/RecSys-Industrial-Book/knowledge-map/ · Source: https://github.com/jhljx/RecSys-Industrial-Book