Self-supervised Auxiliary Learning with Meta-paths for Heterogeneous Graphs

This repository is the implementation of SELAR.

Dasol Hwang^* , Jinyoung Park^* , Sunyoung Kwon, Kyung-min Kim, Jung-Woo Ha, Hyunwoo J. Kim, Self-supervised Auxiliary Learning with Meta-paths for Heterogeneous Graphs, In Advanced in Neural Information Processing Systems (NeurIPS 2020).

Data Preprocessing

We used datasets from KGNN-LS and RippleNet for link prediction. Download meta-paths label (meta_labels/) from this link.

• data/music/

  • ratings_final.npy : preprocessed rating file released by KGNN-LS;
  • kg_final.npy : knowledge graph file;
    • meta_labels/
      • pos_meta{}_{}.pickle : meta-path positive label for auxiliary task
      • neg_meta{}_{}.pickle : meta-path negative label for auxiliary task

• data/book/

  • ratings_final.npy : preprocessed rating file released by RippleNet;
  • kg_final.npy : knowledge graph file;
    • meta_labels/
      • pos_meta{}_{}.pickle : meta-path positive label for auxiliary task
      • neg_meta{}_{}.pickle : meta-path negative label for auxiliary task

Required packages

A list of dependencies will need to be installed in order to run the code. We provide the dependency yaml file (env.yml)

$ conda env create -f env.yml

Running the code

# check optional arguments [-h]
$ python main_music.py
$ python main_book.py

Overview of the results of link prediction

Last-FM (Music)

Book-Crossing (Book)

Citation

@inproceedings{NEURIPS2020_74de5f91,
 author = {Hwang, Dasol and Park, Jinyoung and Kwon, Sunyoung and Kim, KyungMin and Ha, Jung-Woo and Kim, Hyunwoo J},
 booktitle = {Advances in Neural Information Processing Systems},
 editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},
 pages = {10294--10305},
 publisher = {Curran Associates, Inc.},
 title = {Self-supervised Auxiliary Learning with Meta-paths for Heterogeneous Graphs},
 url = {https://proceedings.neurips.cc/paper/2020/file/74de5f915765ea59816e770a8e686f38-Paper.pdf},
 volume = {33},
 year = {2020}
}

License

Copyright (c) 2020-present NAVER Corp. and Korea University