DiseCTR: Disentangled Interest Network for Out-of-Distribution CTR Prediction

This is our TensorFlow implementation for DiseCTR.

The code is tested under a Linux desktop with TensorFlow 2.3 and Python 3.8.10.

Model Training

Unzip the data files in data folder.

Change to the script directory.

cd examples

Use the following command to train a DiseCTR model on Amazon dataset:

python run_disectr.py --dataset amazon --name amazon_test

or on Wechat dataset:

python run_disectr.py --dataset wechat --name wechat_test

or on Kuaishou dataset:

python run_disectr.py --dataset kuaishou --name kuaishou_test

Note

The implemention is based on DeepCTR.