Recommender Systems from Zero
Year: 2026
Type: book
Abstract: Recommender Systems from Zero is a free, open-access handbook that builds modern recommender systems from the ground up — assuming no machine learning, only high-school algebra and the willingness to check a worked example with a calculator. It climbs one vertical path: linear algebra, calculus and probability; losses, neural networks and the Transformer; traditional collaborative filtering and evaluation; graph-based recommenders (LightGCN, contrastive self-supervised learning, the spectral/graph-filter view); the LLM × RecSys landscape (including RLMRec-style alignment); and how to train and serve a recommender. Every concept is introduced before it is used, every formula is decoded component by component, and every worked example is reproducible by hand. Every chapter and appendix was re-read against a written standard and the worked numbers recomputed; dated claims are as of mid-October 2026. Licensed CC BY 4.0.
Authors: Duong, Tan Nghia, Tran, Manh Hoang
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7166319369