This is our Tensorflow implementation for the paper:

Weiyu Cheng, Yanyan Shen, Linpeng Huang. Adaptive Factorization Network: Learning Adaptive-Order Feature Interactions. In AAAI'20, New York, NY, USA, February 07-12, 2020.

Author: Weiyu Cheng (weiyu_cheng at sjtu.edu.cn)

We propose the Adaptive Factorization Network (AFN), a new model that learns arbitrary-order cross features adaptively from data. The core of AFN is a logarithmic transformation layer that converts the power of each feature in a feature combination into the coefficient to be learned. The experimental results on CTR prediction tasks demonstrate the superior predictive performance of AFN against the state-of-the-arts.

If you want to use our codes in your research, please cite:

@inproceedings{DBLP:conf/aaai/ChengSH20,
  author    = {Weiyu Cheng and
               Yanyan Shen and
               Linpeng Huang},
  title     = {Adaptive Factorization Network: Learning Adaptive-Order Feature Interactions},
  booktitle = {The Thirty-Fourth {AAAI} Conference on Artificial Intelligence, {AAAI}
               2020, New York, NY, USA,
               February 7-12, 2020},
  pages     = {3609--3616},
  publisher = {{AAAI} Press},
  year      = {2020},
}

The code has been tested running under Python 3.6. The required packages are as follows:

• tensorflow == 1.10.0

Here we only provide the Frappe and Movielens datasets in this repository due to the large size of the other two datasets. For the Criteo and Avazu datasets, we have uploaded our preprocessed data to codalab. If you'd like to also run experiments on Criteo and Avazu datasets, please first run the downloading script:

cd src
python download_criteo_and_avazu.py

cd src
sh ./run_experiments.sh