Skip to content

An Attention-Based User Behavior Modeling Framework for Recommendation

License

Notifications You must be signed in to change notification settings

jinze1994/ATRank

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ATRank

An Attention-Based User Behavior Modeling Framework for Recommendation

Introduction

This is an implementation of the paper ATRank: An Attention-Based User Behavior Modeling Framework for Recommendation. Chang Zhou, Jinze Bai, Junshuai Song, Xiaofei Liu, Zhengchao Zhao, Xiusi Chen, Jun Gao. AAAI 2018.

Bibtex:

@paper{zhou2018atrank,
  author = {Chang Zhou and Jinze Bai and Junshuai Song and Xiaofei Liu and Zhengchao Zhao and Xiusi Chen and Jun Gao},
  title = {ATRank: An Attention-Based User Behavior Modeling Framework for Recommendation},
  conference = {AAAI Conference on Artificial Intelligence},
  year = {2018}
}

This repository also contains all the competitor's methods mentioned in the paper. Some implementations consults the Transfomer, and Text-CNN.

Note that, the heterogeneous behavior datasets used in the paper is private, so you could not run multi-behavior code directly. But you could run the code on amazon dataset directly and review the heterogeneous behavior code.

Requirements

  • Python >= 3.6.1
  • NumPy >= 1.12.1
  • Pandas >= 0.20.1
  • TensorFlow >= 1.4.0 (Probably earlier version should work too, though I didn't test it)
  • GPU with memory >= 10G

Download dataset and preprocess

  • Step 1: Download the amazon product dataset of electronics category, which has 498,196 products and 7,824,482 records, and extract it to raw_data/ folder.
mkdir raw_data/;
cd utils;
bash 0_download_raw.sh;
  • Step 2: Convert raw data to pandas dataframe, and remap categorical id.
python 1_convert_pd.py;
python 2_remap_id.py

Training and Evaluation

This implementation not only contains the ATRank method, but also provides all the competitors' method, including BPR, CNN, RNN and RNN+Attention. The training procedures of all method is as follows:

  • Step 1: Choose a method and enter the folder.
cd atrank;

Alternatively, you could also run other competitors's methods directly by cd bpr cd cnn cd rnn cd rnn_att, and follow the same instructions below.

Note that, the heterogeneous behavior datasets used in the paper is private, so you could't run the code of this part directly. But you could review the neural network code we use in this paper by cd multi.

  • Step 2: Building the dataset adapted to current method.
python build_dataset.py
  • Step 3: Start training and evaluating using default arguments in background mode.
python train.py >log.txt 2>&1 &
  • Step 4: Check training and evaluating progress.
tail -f log.txt
tensorboard --logdir=save_path

Note that the evaluating producure alternate with training producure, so run the command above may cost five to ten hours until converge completely according to the different methods. If you need to kill that job instantly:

nvidia-smi  # Fetch the PID of current training process.
kill -9 PID # Kill the target process.

You could change the training and networks hyperparameters by command arguments, like python train.py --learning_rate=0.1. To see all command arguments you could use python train.py --help.

Results

You always could use tensorboard --logdir=save_path to see the AUC curve and check all kinds of embedding histogram. The collected AUC curve of test set is as follows

AUC curve in test set

About

An Attention-Based User Behavior Modeling Framework for Recommendation

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published