Jigsaw Toxic Comment Evaluation Set for Content Moderation Models

An evaluation subset of the Jigsaw Toxic Comment Dataset containing Wikipedia talk page comments annotated for toxic behavior. The dataset is hosted by GuardrailsAI and was last updated on February 12, 2025. It is intended for model evaluation, with training recommended from the original Jigsaw dataset.

Use Cases

• Benchmarking toxicity detection models based on human-labeled Wikipedia comments.
• Evaluating model performance on multi-label toxicity classification tasks.
• Testing the robustness of NLP models against toxic language patterns mentioned in the description.

Strengths

• Contains human-annotated labels for six types of toxicity, providing a multi-faceted evaluation target.
• Derived from a well-known, large-scale source dataset of approximately 159,000 Wikipedia comments.
• Explicitly designated for model evaluation, aligning with established machine learning workflows.

Limitations

• Row count, column definitions, and file formats are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Jigsaw Toxic Comment Dataset (Wikipedia comments)

Collection Method

Human raters labeled comments for toxic behavior.

Time Range

null

Freshness

Last updated 2025-02-12 01:53:26; freshness should be verified.

Geography

null