Inf-VAE: A Variational Autoencoder Framework to Integrate Homophily and Influence in Diffusion Prediction

Contributors: Aravind Sankar (asankar3@illinois.edu), Xinyang Zhang (xz43@illinois.edu).

Aravind Sankar, Xinyang Zhang, Adit Krishnan and Jiawei Han, “Inf-VAE: A Variational Autoencoder Framework to
Integrate Homophily and Influence in Diffusion Prediction”, International Conference on Web Search and Data Mining, WSDM 2020, Houston, TX, February 3-7, 2020.

This repository contains a TensorFlow implementation of Inf-VAE: Influence Variational Autoencoder
for information diffusion prediction.
Inf-VAE model covariance of diffusion behaviors with both social network structure and
temporal cascade evolution.
Inf-VAE supports a wide range of graph neural network architectures
as encoders and decoders, to model social homophily.

Note: Though Inf-VAE is designed to model diffusion behaviors in attributed social networks,
our benchmarking experiments are carried out on network datasets that do not contain user attributes.

Input Format

Two stack-exchange datasets Android and Christianity are included in this repository.
In order to use your own data, you have to provide the following files:

• graph.txt: A text file specifying social user links (one pair per line). The first line contains the total numbers of users and links in the social network.

• feats.npz [optional]: A numpy-stored N x D user feature matrix where
  N is the number of users and Dis the number of features per user.

• train.txt, val.txt, test.txt: Text files specifying training, validation, and test cascade sequences (one per line).
  Each line contains a seed user, followed by (user, activation time) pairs.

Requirements

The code has been tested running under Python 3.6 with the following packages installed (along with their dependencies):

tensorflow==1.15.0
numpy==1.18.1
scipy==1.4.1
networkx==2.4
pandas==1.0.1

To guarantee that you have the right package versions, you can use Anaconda to set up a virtual environment and install the above packages.

Repository Organization

• data/ contains the necessary input file(s) for each dataset after pre-processing.

• models/ contains:

  • base python classes to construct layers and models (layers.py, models.py);
  • neural graph autoencoder models MLP + MLP and GCN + Inner Product (graph_ae.py);
  • Inf-VAE model (infvae_models.py);

• utils/ contains:

  • preprocessing subroutines (preprocess.py);
  • hyper-parameter settings (flags.py);
• eval/ contains helper functions to compute evaluation metrics Recall and Mean Average Precision.

Running the Model

To train and evaluate the model (e.g., on android) dataset, please run

python train.py --dataset android --cuda_device 0

Note: The model is not deterministic. All the experimental results provided in the paper are averaged across multiple
runs.

Reference

If you make use of this code or the Inf-VAE algorithm in your work, please cite the following paper:

@inproceedings{sankar2020inf,
  title = {Inf-VAE: A Variational Autoencoder Framework to Integrate Homophily and Influence in Diffusion Prediction},
  author = {Sankar, Aravind and Zhang, Xinyang and Krishnan, Adit and Han, Jiawei},
  doi = {10.1145/3336191.3371811},
  booktitle = {Proceedings of the 13th International Conference on Web Search and Data Mining},  
  pages = {510--518},
  year = {2020},
}