Civil Comments: 2 Million Public News-Site Comments with Crowd-Rated Toxicity and Identity-Mention Labels (Jigsaw, Google)
- Provider
- Jigsaw and Google (Borkan et al.); Civil Comments platform
- License
- CC0-1.0 (source)
- Size
- 6 files
- Format
- Parquet (in sampled files)
- Stored at
- Hugging Face (https://huggingface.co/datasets/google/civil_comments)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Natural Language Processing
- Source
- https://arxiv.org/abs/1903.04561
About this dataset
Civil Comments holds about two million public comments posted between 2015 and 2017 on roughly 50 English-language news sites that used the Civil Comments commenting plugin. When the platform shut down in 2017 it released the comment text as an open archive for research, and Jigsaw (Google) added crowd-sourced labels: a toxicity score for each comment (the fraction of raters who judged it toxic) plus sub-attributes such as severe toxicity, obscene, threat, insult, identity attack and sexually explicit. This is the data released for the Jigsaw Unintended Bias in Toxicity Classification challenge, widely used to study hate speech and abusive language detection, online harassment, content moderation and bias against identity groups in text classifiers. Splits: 1,804,874 training, 97,320 validation and 97,320 test comments, stored as four Parquet files (about 420 MB) under data/. Load with the Hugging Face datasets library (load_dataset('google/civil_comments')) or pandas. Released under CC0 1.0, as is the underlying comment text; cite Borkan et al., Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classification, WWW 2019 Companion.
How to load this dataset in Python
from datasets import load_dataset
ds = load_dataset("google/civil_comments")
print(ds)Data structure
File types in sampled files: Parquet
Files
Example file: data/test-00000-of-00001.parquet
Analyze this dataset without downloading it
On Scigantic, Civil Comments: 2 Million Public News-Site Comments with Crowd-Rated Toxicity and Identity-Mention Labels (Jigsaw, Google) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Natural Language Processing datasets
- Community Alignment Dataset
- Nemotron RL Instruction Following MultiTurnChat V 1
- WikiDomains: 22,561 Wikipedia Term Definitions across 13 Academic Domains with Annotated Difficult Concepts for Targeted Concept Simplification (Google DeepMind)
- Common Crawl
- Japanese Tokenizer Dictionaries
- SMOL and GATITOS: Professional Translations into 221 Low-Resource Languages (Google Research)
- Wiki Dpr
- CantTalkAboutThis Topic Control Dataset NC