Recommender Systems: Industrial Architecture and Core Algorithms
An in-progress Chinese-language technical book about industrial recommendation-system architecture and algorithms, covering the pipeline from recall through ranking, mixing and generative recommendation, for engineers with a machine-learning background.
This book manuscript covers industrial recommendation-system design end to end: system architecture, evaluation methodology, recall and filtering, coarse and fine ranking, re-ranking and blending, and generative recommendation modules, aimed at graduate students, algorithm and architecture engineers, and researchers who already have a machine-learning or recommendation background rather than complete beginners.
Completed chapters, released as PDFs with runnable companion code and experiment cases, cover an overview of recommendation systems, system architecture and evaluation, recall models including collaborative filtering, graph recall, vector recall and two-tower models, coarse-ranking models, feature engineering, offline serving infrastructure, vector-retrieval systems, systematic bias problems in recommendation, and watch-duration estimation modeling. The author reports the upper volume about 75% complete and the full two-volume work about 50% complete, roughly 600 pages so far, with remaining chapters on user-interest modeling, multi-objective optimization, reinforcement learning and LLM- or agent-based recommendation still planned. It is distributed for personal, non-commercial use.