Cora
Task: Paper topic classification (citation network)
Size band: small · Label type: semantic
Label column: subject · Converter: converters/convert_linqs.py
Classic citation network of machine-learning papers; the label is one of 7 subject areas and features are 1,433 binary bag-of-words indicators. Strongly homophilous — useful as a calibration baseline against the GNN literature.
Converted graphs (neext/)
| graph | nodes | edges | classes | feature cols | isolated | class counts |
|---|---|---|---|---|---|---|
| default | 2,708 | 5,278 | 7 | 1433 | 0 | Neural_Networks: 818, Probabilistic_Methods: 426, Genetic_Algorithms: 418, Theory: 351, Case_Based: 298, Reinforcement_Learning: 217, … |
Conversion notes: LINQS .content/.cites; 1433 binary bag-of-words feature columns; 0 citation edges dropped (endpoint not in .content); source paper IDs remapped (id_mapping.csv).
Source
- cora.tgz — 168,052 bytes, sha256
0d4ed463d1627bb7…, fetched 2026-07-23
License: LINQS research distribution
Citation: Sen, Namata, Bilgic, Getoor, Gallagher, Eliassi-Rad. Collective Classification in Network Data. AI Magazine 2008.
Caveats
- Homophily-driven labels favor message-passing GNNs; egonet embeddings are expected to trail here.
Generated by converters/make_cards.py; stats from metadata.json. Raw files: source/. NEExT tables: neext/.