https://github.com/cgcl-codes/ehgnn
An implementation for the paper--Efficient Learning for Billion-scale Heterogeneous Information Networks.
Science Score: 13.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
○CITATION.cff file
-
✓codemeta.json file
Found codemeta.json file -
○.zenodo.json file
-
○DOI references
-
○Academic publication links
-
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (4.6%) to scientific vocabulary
Repository
An implementation for the paper--Efficient Learning for Billion-scale Heterogeneous Information Networks.
Basic Info
Statistics
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
- Releases: 0
Metadata Files
README.md
EHGNN
An implementation for the paper--Efficient Learning for Billion-scale Heterogeneous Information Networks.
Dataset
The three datasets used in the paper (PubMed, Yelp and DBLP) can be downloaded from here. In addition, the OGB-MAG240M dataset can be found here. Please place the downloaded datasets in the ../data.
Usage
To conduct the experiments, please execute main.py in each folder (Node Classification, Link Prediction, and MAG240M). Hyperparameters can be explored within the main.py, and here are the ones we used.
| Task | Dataset | $\alpha$ | K | learning rate | dropout | hidden dimension | layers | batch size | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Node Classification | PubMed | 0.7 | 20 | 1e-3 | 0.4 | 256 | 4 | 3000 | | | Yelp | 0.7 | 20 | 3e-4 | 0.5 | 256 | 4 | 3000 | | | DBLP | 0.7 | 20 | 5e-4 | 0.5 | 512 | 5 | 3000 | | | OGB-MAG240M | 0.7 | 25 | 3e-4 | 0.4 | 512 | 5 | 5000 | | Link Prediction | PubMed | 0.1 | 20 | 3e-4 | 0.5 | 256 | 4 | 40 | | | Yelp | 0.1 | 20 | 3e-4 | 0.5 | 256 | 4 | 100 | | | DBLP | 0.7 | 20 | 5e-4 | 0.5 | 512 | 5 | 1000 |
Owner
- Name: CGCL-codes
- Login: CGCL-codes
- Kind: organization
- Website: http://grid.hust.edu.cn/
- Repositories: 35
- Profile: https://github.com/CGCL-codes
CGCL/SCTS/BDTS Lab