Twitch Social Networks (6 languages)
Task: Explicit-content streamer classification
Size band: small-medium · Label type: semantic
Label column: mature · Converter: converters/convert_musae.py
Six independent friendship networks of Twitch streamers (DE, ENGB, ES, FR, PTBR, RU), 1.9k-9.5k nodes each. The binary label flags streamers who use explicit language. Six graphs from one source enable cross-graph generalization studies (train on one language, test on another).
Converted graphs (neext/)
| graph | nodes | edges | classes | feature cols | isolated | class counts |
|---|---|---|---|---|---|---|
| de | 9,498 | 153,138 | 2 | 0 | 0 | 1: 5,742, 0: 3,756 |
| engb | 7,126 | 35,324 | 2 | 0 | 0 | 1: 3,888, 0: 3,238 |
| es | 4,648 | 59,382 | 2 | 0 | 0 | 0: 3,288, 1: 1,360 |
| fr | 6,549 | 112,666 | 2 | 0 | 0 | 0: 4,135, 1: 2,414 |
| ptbr | 1,912 | 31,299 | 2 | 0 | 0 | 0: 1,251, 1: 661 |
| ru | 4,385 | 37,304 | 2 | 0 | 0 | 0: 3,310, 1: 1,075 |
Conversion notes: Six independent language graphs; 'new_id' is the graph node id; label = mature (explicit-content flag, 0/1). Feature JSONs (games/habits token lists) not columnized.
Source
- twitch.zip — 2,842,994 bytes, sha256
65a6c4c23da23889…, fetched 2026-07-23
License: SNAP / MUSAE; GPL-3.0 code, cite MUSAE
Citation: Rozemberczki, Allen, Sarkar. Multi-Scale Attributed Node Embedding (MUSAE). Journal of Complex Networks 2021.
Caveats
- Feature JSONs (games/habits token lists) not columnized; FR listed two accounts twice (deduped, labels agreed).
Generated by converters/make_cards.py; stats from metadata.json. Raw files: source/. NEExT tables: neext/.