NEExT / webkb /card.md
anomalypoint's picture
Add files using upload-large-folder tool
3a12cf2 verified
|
Raw
History Blame Contribute Delete
2.65 kB

WebKB (Cornell / Texas / Wisconsin)

Task: University web-page classification
Size band: small · Label type: semantic
Label column: page_class · Converter: converters/convert_geomgcn.py

Three tiny university web-page networks with 5 page classes and 1,703 binary word features. Strongly heterophilous — classic stress tests where neighbor labels mislead.

Converted graphs (neext/)

graph nodes edges classes feature cols isolated class counts
cornell 183 277 5 1703 0 3: 82, 0: 38, 2: 30, 4: 17, 1: 16
texas 183 279 5 1703 0 3: 101, 0: 33, 4: 30, 2: 18, 1: 1
wisconsin 251 450 5 1703 0 2: 118, 1: 70, 3: 32, 4: 21, 0: 10

Conversion notes: geom-gcn out1 files; 5 page classes (course/faculty/student/project/staff as ints); 1703 binary bag-of-words f_* columns; directed web links symmetrized.

Source

License: Not stated (geom-gcn repo)
Citation: Pei, Wei, Chang, Lei, Yang. Geom-GCN: Geometric Graph Convolutional Networks. ICLR 2020 (data packaging).

Caveats

  • Very small graphs (183-251 nodes); high result variance — use many splits.

Generated by converters/make_cards.py; stats from metadata.json. Raw files: source/. NEExT tables: neext/.