This repository include a solution for HaSpeeDe3 challenge, focuesed on Hate Speech Detection in Twitter data using binary classification models. The tasks include both in-domain detection (within the same dataset) and cross-domain detection (across different datasets). Participants can use textual data alone or combine it with contextual metadata for improved accuracy.
Goal: Classify whether tweets contain political hate speech.
- Sub-tasks:
- Textual: Use only the tweet content from the PolicyCorpusXL dataset.
- Contextual: Use tweet content + additional metadata (e.g., author details, friends, replies, retweets).
Goal: Detect hate speech across different domains.
- Sub-tasks:
- XPoliticalHate: Test on political hate speech (PolicyCorpusXL).
- XReligiousHate: Test on religious hate speech using the ReligiousHate dataset.
Participants can optionally incorporate external data sources for Task B.
- PolicyCorpusXL: Dataset for political hate speech tasks.
- ReligiousHate: Dataset for religious hate speech detection.
| Team | Task A (textual/contextual) | Task B (XPoliticalHate) | Task B (XReligiousHate) |
|---|---|---|---|
| our team | 0.9192 | 0.9192 | 0.6476 |
| BERTicelli | 0.8976 | 0.8976 | 0.5401 |
| CHILab | 0.8516 | 0.8516 | - |
| extremITA | 0.9079 | 0.9079 | 0.6525 |
| INGEOTEC | 0.8845 | 0.8845 | 0.5522 |
| LMU | - | 0.9014 | 0.6461 |
| odang4 | 0.9128 | 0.9128 | 0.5213 |