HandyRec is a package for deep-learning recommendation models implemented with TF2.6 ✨. It is meant to be an easy-to-use and easy-to-read package for people who want to use or learn classic deep-learning recommendation models.
It is currently a personal project for learning purposes. I recently started to learn deep-learning recommendation algorithms and design patterns💦. I'll try to implement some classical algorithms along with example notebooks here.
| Model | Paper | Example |
|---|---|---|
| YouTubeDNN | [RecSys 2016] Deep Neural Networks for YouTube Recommendations |  |
| DSSM | [CIKM 2013] Learning Deep Structured Semantic Models for Web Search using Clickthrough Data | |
| Model | Paper | Example |
|---|---|---|
| YouTubeDNN | [RecSys 2016] Deep Neural Networks for YouTube Recommendations | |
| DeepFM | [IJCAI, 2017] DeepFM: A Factorization-Machine based Neural Network for CTR Prediction | |
| Model | Paper | Example |
|---|---|---|
| DIN | [SIGKDD, 2018] Deep Interest Network for Click-Through Rate Prediction | |
| DIEN | [AAAI, 2019] Deep Interest Evolution Network for Click-Through Rate Prediction | |
| FMLP-Rec | [WWW, 2022] Filter-enhanced MLP is All You Need for Sequential Recommendation | |
The main usage flow is shown below:
For more details, examples can be found here and the table above. Documentation can be found here.
NOTE: This project is under development and has not been packaged yet😣. Please download the source code and import it as a local module. 🚧 I'll package this project when >10 models are implemented.
Especially thanks to DeepMatch and DeepCTR. I got much inspiration about code structure and model implementation from these projects.
The logo of this project is inspired by AdobeLogoMaker.