Dataset Overview

This section provides detailed documentation for all supported datasets in PyGDA, including their domains and sources.

Citation Networks

Arxiv

• Domains: 3 domains based on publication years
• Source: ogbn-arxiv
• Processing: See ArxivDataset
• Features: Generated from paper abstracts
• Note: Can be preprocessed with scripts in benchmark folder

Citation

• Domains: ACMv9, Citationv1, DBLPv7
• Source: Adopted from ASN
• Processing: See CitationDataset
• Download: Download Link

MAG

• Domains: CN, DE, FR, JP, RU, US
• Source: Originally from ogbn-mag
• Processing: See MAGDataset
• Download: Download Link
• Note: Separated into 6 countries by PairAlign

Social Networks

Blog

• Domains: Blog1, Blog2
• Source: Adopted from ACDNE
• Processing: See BlogDataset
• Download: Download Link

Twitch

• Domains: DE, EN, ES, FR, PT, RU
• Source: Twitch Social Networks
• Processing: See TwitchDataset
• Download: Download Link

Infrastructure Networks

Airport

• Domains: Brazil, Europe, USA
• Source: Adopted from struc2vec
• Processing: See AirportDataset
• Features: Constructed using OneHotDegree for each node
• Download: Download Link

Graph Classification Benchmarks

TUGraph

• Datasets:

  • PROTEINS
  • FRANKENSTEIN
  • Mutagenicity

• Domains: 2 domains based on density for each dataset

• Source: Adopted from TUDataset
• Processing: See GraphTUDataset
• Download: Download Link

Usage Example

from pygda.datasets import CitationDataset

# Load the Citation dataset
dataset = CitationDataset(root='data/citation', name='ACMv9')
data = dataset[0]

# Access the data
x = data.x  # Node features
edge_index = data.edge_index  # Graph connectivity
y = data.y  # Labels

Each dataset documentation includes:

• Detailed domain descriptions
• Data sources and references
• Processing instructions
• Download information
• Usage examples
• Implementation details

For specific details about each dataset, please visit their respective documentation pages linked above.