📚 Notes on RecSys Models

My notes on models in recommendation systems. All models are implemented in PyTorch, with simple usage demos and brief introduction markdown notes.

Welcome to Notes-on-RecSys-Models! This repository is a collection of my learning notes and PyTorch implementations of classic and state-of-the-art recommendation system algorithms.

The goal of this project is to provide:

1. Clean Code: Readable PyTorch implementations focused on the model structure.
2. Theory: Brief markdown notes explaining the core idea behind each model.
3. Practice: Simple demos to show how to input data and get predictions.

📂 Directory Structure

Here is the organization of the project:

Notes-on-RecSys-Models/
├── assets/                 # Images and diagrams for documentation
├── Classic Models/         # Model implementations organized by stage
│   ├── Recall/             # Retrieval models
│   ├── Ranking/            # Fine-ranking models
│   └── Re-ranking/         # Re-ranking strategies
├── Data/                   # Raw data storage (e.g., Amazon Review 2018)
├── Datasets/               # PyTorch Dataset wrappers and preprocessing logic
│   └── BaseDatasets.py     # Base classes for datasets
├── Utils/                  # Common utility functions (I/O, metrics)
└── README.md

🚀 Implemented Models

Models are organized by the stage they are typically used in a recommendation pipeline: Recall (Retrieval), Ranking (Fine-sorting), and Re-ranking.

1. Recall

Goal: Retrieve a small set of relevant items from a massive pool of candidates efficiently.

Model	Paper / Source	Code	Note	Status
YoutubeDNN	Deep Neural Networks for YouTube Recommendations	Link	📝	✅
MIND	Multi-Interest Network with Dynamic Routing for Recommendation at Tmall	Link	📝	✅
SDM	SDM: Sequential Deep Matching Model for Online Large-scale Recommender System	Link	📝	🚧
HSTU	Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Generative Recommendations	❌	📝	✅
TIGER	Recommender Systems with Generative Retrieval	❌	📝	✅

2. Ranking

Goal: Accurately predict user preference (CTR/CVR) for the retrieved items to sort them.

Model	Paper / Source	Code	Note	Status

3. Re-ranking

Goal: Optimize the final list for diversity, novelty, or business constraints.

Model	Source / Concept	Code	Note	Status

(Legend: ✅ Implemented, 🚧 In Progress, 📅 Planned)