MAP-CODE

Official Code for paper "MAP: A Model-agnostic Pretraining Framework for Click-through Rate Prediction"

NOTE: I have deleted some unrelated codes which are our preliminary exploratory experiments. If readers got any problems or come across any bugs, please kindly leave me a message.

Requirement

pip install -r requirements.txt

Data Preprocessing

We provide the data preprocessing scripts in data_preprocess folder. One can also download the preprocessed data from [Link] and place it at the main folder.

Quick Start

We provide demo scripts in run_script folder.

To train DCNv2 from scratch:

CUDA_VISIBLE_DEVICES=0 bash run_script/run_DCNv2_scratch.sh

To pretrain DCNv2 with MFP:

CUDA_VISIBLE_DEVICES=0 bash run_script/run_DCNv2_MFP.sh

To pretrain DCNv2 with RFD:

CUDA_VISIBLE_DEVICES=0 bash run_script/run_DCNv2_RFD.sh

To finetune DCNv2 after pretraining:

CUDA_VISIBLE_DEVICES=0 bash run_script/run_DCNv2_finetune.sh

Citation

@inproceedings{lin2023map,
  title={MAP: A Model-agnostic Pretraining Framework for Click-through Rate Prediction},
  author={Lin, Jianghao and Qu, Yanru and Guo, Wei and Dai, Xinyi and Tang, Ruiming and Yu, Yong and Zhang, Weinan},
  booktitle={Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
  pages={1384--1395},
  year={2023}
}