Skip to content

celi52/si-hdgnn

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

30 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SI-HDGNN: Heterogeneous Dynamical Academic Network for Learning Scientific Impact Propagation

This repo provides a reference implementation of SI-HDGNN.

Heterogeneous Dynamical Academic Network for Learning Scientific Impact Propagation
Xovee Xu, Ting Zhong, Ce Li, Goce Trajcevski, and Fan Zhou
Knowledge-Based Systems, vol. 238, pp. 107839, Feb 2022

Requirements

The code was tested with Python 3.7, tensorflow-gpu 2.4, torch 1.8.1, cudnn 8.0 and cudatoolkit 11.0. Install the dependencies via Anaconda:

# create conda virtual environment
conda create --name si-hdgnn -c conda-forge cudatoolkit=11.0 cudnn=8.0

# activate environment
conda activate si-hdgnn

# install other dependencies
pip install -r requirements.txt

Hint: pay attention to the versions of cudatoolkit and cudnn, tensorflow and torch rely on certain versions of them for GPU/TPU acceleration.

Datasets

APS and its preprocessd data can be downloaded in Google Drive.

You can access the original APS dataset here. (Released by American Physical Society, obtained at Jan 17, 2019)

Or DBLP-Citation-network V10, and ACM-Citation-network V9 here. (Released by Aminer)

Run the code

For a given scientific dataset, you should:

  1. Construct a heterogeneous graph
  2. Get node embeddings
  3. Generate scientific information cascades
  4. Training & Evaluating

Detailed pre-process files information can be found here.

1. Construct heterogeneous graph

This stage may costs a large amount of RAM (~64GB with millions of nodes/edges in graph).

# build a heterogeneous graph
> python codes/gnn_pre/graph_sample.py

# heterogeneous neighboring node sampling save and run
> python codes/gnn_pre/save_rwr.py
> python codes/gnn_pre/run_rwr.py

2. Generate node embeddings

After graph construction, we now learn node embeddings via a heterogeneous graph neural network.

> python codes/gnn_train/pre_train_files.py 
> python codes/gnn_train/gene_node_embeddings.py --train_iter_n 30

3. Generate scientific information cascades

Once we got the node embeddings, we can generate cascades and corresponding training/validation/test data.

Paper Prediction Run scripts:

> python codes/predict_paper/1_load_emb.py
> python codes/predict_paper/2_construct_cascade.py

Author Prediction Run scripts:

> python codes/predict_author/1_load_emb.py
> python codes/predict_author/2_x_y.py

4. Training & Evaluating SI-HDGNN

> python codes/predict_paper/paper_prediction.py
> python codes/predict_author/author_prediction.py

Options

You may change the model settings manually in config.py or directly into the codes.

Cite

If you find SI-HDGNN useful for your research, please consider citing us 😘 :)

@article{xu2021heterogeneous, 
  title = {Heterogeneous Dynamical Academic Network for Learning Scientific Impact Propagation}, 
  author = {Xovee Xu and Ting Zhong and Ce Li and Goce Trajcevski and Fan Zhou}, 
  journal = {Knowledge-Based Systems}, 
  year = {2022}, 
  numpages = {20},
  issue = {238}, 
  pages = {107839},
}

Contact

If you have any questions, feel free to contact us, emails: xovee@ieee.org or ce.lc@outlook.com.

About

SI-HDGNN: Heterogeneous Dynamical Academic Network for Scientific Impact Propagation Learning

Topics

Resources

License

Stars

Watchers

Forks

Languages