# Weibo (PyGOD)

**Task**: Social spam detection  
**Size band**: medium · **Label type**: anomaly  
**Label column**: `is_outlier` · **Converter**: `converters/convert_pygod.py`

Sina Weibo user-user graph (shared-hashtag edges) with 400 post-derived features and organic spammer labels (10.3% anomalous). Real social misbehavior at Workbench-comfortable scale, but dense (avg degree ~90).

## Converted graphs (neext/)

| graph | nodes | edges | classes | feature cols | isolated | class counts |
|---|---|---|---|---|---|---|
| default | 8,405 | 377,271 | 2 | 400 | 0 | 0: 8,058, 1: 347 |

*Conversion notes*: PyGOD .pt (torch pickle, trusted pygod-team/data source) via stub unpickling; edge_index symmetrized; 400 x_* feature columns.

## Source

- [weibo.pt.zip](https://raw.githubusercontent.com/pygod-team/data/main/weibo.pt.zip) — 13,339,024 bytes, sha256 `c4a5fe4ca61a9566…`, fetched 2026-07-23

**License**: MIT (pygod-team/data)  
**Citation**: Liu et al. BOND. NeurIPS 2022 D&B (data: Zhao et al.).

- https://github.com/pygod-team/data

## Caveats

- Dense — k_hop=1 with sampling first.
- Torch-pickle source; stub-unpickled from the hash-verified official repo.

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