Dataset · google
News comment archive from news sites for moderation
google/civil_comments
Millions of English comments from 50 news sites for training text classification and moderation systems.
What's inside
The dataset contains between 1 and 10 million public English comments in text and tabular format. The data is taken from the Civil Comments platform archive and was left by users on approximately 50 independent news sites from 2015 to 2017.
What you can do with it
- Training classifiers for automatic comment moderation and filtering
- Sentiment analysis and user discussions in news media
- Pre-training or fine-tuning text models on conversational vocabulary
- Tasks
- Text classification
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 19K
- Likes
- 46
Models trained on it
- StrandsAgents/strands-decider-2B-qwen3.5-v1-2610Text classification 25K
- StrandsAgents/strands-decider-2B-hobson-v21Text classification 15.3K
- StrandsAgents/strands-decider-2B-hobson-v19Text classification 7K
- Horizon-Labs/multilingual-toxicity-baseText classification 1.6K
- StrandsAgents/strands-decider-E4B-gemma4-v1-2610Text classification 1.4K
- Horizon-Labs/multilingual-toxicity-smallText classification 1.3K
- Horizon-Labs/content-safety-guard-smallText classification 1.1K
- StrandsAgents/strands-decider-12B-gemma4-v1-2610Text classification 1.1K
How to load it
from datasets import load_dataset
ds = load_dataset("google/civil_comments")