README.md

metadata

license: cc-by-sa-4.0
task_categories:
  - text-classification
language:
  - en
tags:
  - hate-speech
  - social-media
  - intent-detection
  - impact-analysis
  - content-moderation
size_categories:
  - 1K<n<10K

I2-HATE: Intent and Impact Hate Speech Dataset

Dataset Description

The I2-HATE dataset introduces a novel dual-taxonomy approach to hate speech detection that separately captures Intent (the speaker's underlying motivations) and Impact (potential societal consequences). Unlike traditional hate speech datasets that use simple categorical labels, I2-HATE enables more nuanced content moderation by distinguishing between why hate speech is produced and what harm it may cause.

This dataset contains 3,296 annotated social media posts with multi-label annotations across 7 intent categories and 8 impact categories.

Key Features

• Size: 3,296 samples
• Multi-label annotations: Each post can have multiple intent and impact labels
• Dual taxonomy framework: Separate classification of Intent and Impact

Intent Labels (7 categories)

1. Affective Aggression [AA]: Emotional outbursts and aggressive expressions
2. Derisive Trolling [DT]: Mocking and ridicule intended to provoke
3. Dominance & Subjugation [D&S]: Assertions of superiority and control
4. Ideological Expression [IE]: Promotion of specific ideological beliefs
5. Performative Reinforcement [PR]: Public displays to reinforce group identity
6. Strategic Incitement [SI]: Deliberate attempts to mobilize others
7. Threat & Intimidation [T&I]: Direct or implied threats

Impact Labels (8 categories)

1. Disruption of Public Discourse [DPD]: Undermining constructive dialogue
2. Glorification of Hate [GH]: Celebrating hateful acts or ideologies
3. Incitement to Discrimination/Exclusion [ID/E]: Encouraging discriminatory behavior
4. Incitement to Violence [IV]: Promoting violent actions
5. Misinformation/Disinformation Nexus [M/DN]: Spreading false narratives
6. Normalization of Prejudice [NP]: Making prejudice socially acceptable
7. Psychological Harm [PH]: Causing emotional or mental distress
8. Stigmatization & Dehumanization [S&D]: Devaluing individuals or groups

Dataset Structure

Each sample contains:

• sample_id: Unique identifier (integer)
• text: The social media post text
• Intent Labels: Comma-separated intent categories
• Impact Labels: Comma-separated impact categories

Citation

If you use the I2-HATE dataset in your research, please cite our paper:

@inproceedings{singhal2026wordswear,
    title={When Words Wear Masks: Detecting Malicious Intents and Hostile Impacts of Online Hate Speech},
    author={Singhal, Priyansh and Joshi, Piyush},
    booktitle={Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 2: Short Papers)},
    year={2026},
    publisher={Association for Computational Linguistics}
}

License

This dataset is released under CC BY-SA 4.0 (Creative Commons Attribution-ShareAlike 4.0 International). You are free to:

• Share and redistribute the dataset
• Adapt and build upon the dataset

Under the following terms:

• Attribution: You must give appropriate credit by citing our paper
• ShareAlike: If you remix, transform, or build upon the dataset, you must distribute your contributions under the same CC BY-SA 4.0 license

Contact

For questions or issues regarding the dataset, please contact:

• Priyansh Singhal

Ethical Considerations

This dataset contains real social media posts with hate speech content. Researchers using this dataset should:

• Handle the data responsibly and ethically
• Consider potential biases in annotation
• Use the dataset solely for research purposes to combat online hate speech