MIRRN

KDD2025 ! ! !

This repository is the implementation for Paper "Multi-granularity Interest Retrieval and Refinement Network for Long-Term User Behavior Modeling in CTR Prediction".

Requirements

• Ensure you have Python and PyTorch (version 1.8 or higher) installed. Our setup utilized Python 3.8 and PyTorch 1.13.0.
• Should you wish to leverage GPU processing, please install CUDA.

Dataset

We use three public real-world datasets (Taobao, Alipay and Tmall) in our experiments. We pre-process the data in the same way with ETA and SDIM. You can download the datasets from the links below.

• Taobao: The raw dataset can be downloaded from https://tianchi.aliyun.com/dataset/dataDetail?dataId=649. If you want to know how to preprocess the data, please refer to ./data/taobao/preprocess.py
• Alipay: The raw dataset can be downloaded from https://tianchi.aliyun.com/dataset/dataDetail?dataId=53. If you want to know how to preprocess the data, please refer to ./data/alipay/preprocess.py
• Tmall: The raw dataset can be downloaded from https://tianchi.aliyun.com/dataset/dataDetail?dataId=42. If you want to know how to preprocess the data, please refer to ./data/tmall/preprocess.py

Example

If you have downloaded the source codes, you can train MIRRN model.

$ cd main
$ python build_taobao_to_parquet.py
$ python run_expid.py

You can change the model parameters in ./main/config/General_config/model_config.yaml

Contact

Should you have any questions regarding our paper or codes, please don't hesitate to reach out via email at demon@mail.ustc.edu.cn.

Acknowledgment

Our code is developed based on reczoo/FuxiCTR: A configurable, tunable, and reproducible library for CTR prediction https://fuxictr.github.io.