Charles-Cao/

TokenAD

Datasets for the paper Towards Token-Level Text Anomaly Detection. Each dataset provides both document-level and token-level anomaly annotations, enabling fine-grained localization of anomalies within text rather than only flagging whole documents.

	Datasets




	

Dataset Anomaly type Documents Anomalies Anomaly rate

SMS Spam Character-level (corrupted / gibberish sequences) 4,518 393 8.7%

Restaurant… See the full description on the dataset page: https://huggingface.co/datasets/Charles-Cao/TokenAD.