14,100 tweets annotated via crowdsourcing for offensive language identification. The dataset was created by christophsonntag and released with the paper 'Predicting the Type and Target of Offensive Posts in Social Media'.
Use Cases
- Train a classifier to predict the subcategory of offensive language from tweet text.
- Analyze the distribution of offensive language types across the 14,100 annotated tweets.
- Build a model to identify the target of offensive posts using the provided annotation labels.
Strengths
- Contains 14,100 annotated data points.
- Annotations include three subcategories for detailed offensive language analysis.
- Focuses on offensive language as a whole, not just specific types like hate speech.
Limitations
- Limited to tweets from a single platform, which may not generalize to other social media or text formats.
- Annotations are from crowdsourcing, which can introduce label noise or inconsistency.
- Dataset size of 14,100 rows may be insufficient for training very large language models without augmentation.
Provenance
- Source
- Twitter tweets.
- Collection Method
- Crowdsourced annotation.
- Time Range
- null
- Freshness
- Last updated on 2024-03 15.
- Geography
- null