fanoping / DIN-pytorch

PyTorch Implementation of Deep Interest Network for Click-Through Rate Prediction

PyTorch Implementation of Deep Interest Network for Click-Through Rate Prediction

Overview

This repository contains code unofficially reimplemented the original paper KDD 2018 Deep Interest Network for Click-Through Rate Prediction in PyTorch version

To-Dos

• Simple Architecture Reimplementation
• Dice Activation / PReLU
• Data Evaluation (not trained yet)
• Dropout
• Regularization in DiFacto
• Mini-Batch Aware (MBA) regularization

Acknowledgements

Part of codes are adapted from the following projects

• Deep Interest Network: Official repository for DIN

• DeepCTR: A collection of Click-Through Rate (CTR) prediction works