Datasets:

JaySuryavanshi
/

graph-anomaly-weibo

license: mit
configs:
  - config_name: nodes
    default: true
    data_files:
      - split: train
        path: nodes.parquet
  - config_name: edges
    data_files:
      - split: train
        path: edges.parquet
task_categories:
  - graph-ml
tags:
  - anomaly-detection
  - graph-anomaly-detection
  - outlier-detection
  - fraud-detection
  - graph
pretty_name: Weibo (graph anomaly detection)
size_categories:
  - 10K<n<100K

Weibo (graph anomaly detection)

Weibo users connected by shared hashtag activity, with labeled suspicious accounts. An organic (non-injected) anomaly dataset used throughout the BOND benchmark line of work.


Nodes	8,405
Node features	400
Edges	407,963
Outliers	347 (4.1%)
Label type	proxy
Label source	suspicious-account labels via the BOND benchmark

Viewer. Two configs: nodes (default) and edges.

Files. nodes.parquet (node_id, feat_*, label, split masks where available) and edges.parquet (src, dst, the edge list as shipped upstream; symmetrize for undirected use). Ships the upstream single split as train_mask_0, val_mask_0, test_mask_0.

A provenance note this card refuses to paper over: the upstream README table lists 868 outliers (10.3%) for Weibo, but the shipped weibo.pt file contains 347 (4.1%). The numbers on this card describe the actual file (sha256 of the source zip: c4a5fe4c...). If you compare against papers reporting the 10.3% variant, you are not using the same labels.

Load

import pandas as pd

nodes = pd.read_parquet("hf://datasets/JaySuryavanshi/graph-anomaly-weibo/nodes.parquet")
edges = pd.read_parquet("hf://datasets/JaySuryavanshi/graph-anomaly-weibo/edges.parquet")

As a graph, with graphspot (pip install graphspot):

import numpy as np, scipy.sparse as sp, graphspot
from graphspot.detectors import XGBGraph

n = len(nodes)
adj = sp.csr_matrix((np.ones(len(edges)), (edges.src, edges.dst)), shape=(n, n))
g = graphspot.Graph(adj=adj, x=nodes.filter(like="feat_").to_numpy(),
                    node_labels=nodes.label.to_numpy())

Provenance

Mirrored unmodified (beyond format conversion to parquet) from pygod-team/data (MIT). Conversion is scripted and deterministic; label counts above are computed from the files in this repository, not copied from upstream docs.

Citation

Paper: https://arxiv.org/abs/2206.10071

@inproceedings{liu2022bond,
  title={{BOND}: Benchmarking unsupervised outlier node detection on static attributed graphs},
  author={Liu, Kay and Dou, Yingtong and Zhao, Yue and Ding, Xueying and Hu, Xiyang and Zhang, Ruitong and Ding, Kaize and Chen, Canyu and Peng, Hao and Shu, Kai and Sun, Lichao and Li, Jundong and Chen, George H. and Jia, Zhihao and Yu, Philip S.},
  booktitle={Advances in Neural Information Processing Systems},
  year={2022}
}