HATEDEMICS Models and Data

This repository contains resources developed within the HATEDEMICS project, including models and annotated datasets for hate speech detection in English, Spanish, Polish, and Italian.

Contents

• Final versions of hate speech detection models
• Human-annotated datasets
• LLM-annotated datasets
• Documentation on labels, data format, and model usage

Languages

The repository currently includes resources for:

• English
• Spanish
• Polish
• Italian

Tasks

Hate Speech Detection

Models and annotations for identifying hate speech in online content.

Models

The released models were developed using MaChAmp. We provide final models for hate speech detection in English, Spanish, Polish, and Italian.

The released models are fine-tuned versions of existing pre-trained models. Each model was fine-tuned on the corresponding annotated data released in the data/ directory. The MaChAmp configuration files used for training are provided in the configs/ directory.

Released Models

Task	Language	Base model	Training data
Hate speech	Italian	MilaNLProc/hate-ita	Human-annotated Italian Telegram data
Hate speech	Polish	ptaszynski/bert-base-polish-cyberbullying	Human-annotated Polish Telegram data
Hate speech	English	facebook/roberta-hate-speech-dynabench-r4-target	LLM-annotated English Telegram data
Hate speech	Spanish	dccuchile/bert-base-spanish-wwm-uncased	LLM-annotated Spanish Telegram data

Repository Organization

The repository is organized around the three main types of released resources: annotated data, fine-tuned models, and MaChAmp configuration files.

.
├── data/
│   ├── human_annotated/
│   └── llm_annotated/
├── models/
│   └── hate_speech/
├── configs/
│   ├── datasets/
│   └── parameters/
├── docs/
└── README.md

Related Resources

This repository is connected to the MuLTa-Telegram resource, a publicly available Italian and Polish Telegram dataset for hate speech and target detection.

The MuLTa-Telegram dataset is described in the following paper:

• MuLTa-Telegram: A Fine-Grained Italian and Polish Dataset for Hate Speech and Target Detection

Acknowledgements

This work was supported by the European Union’s CERV fund under Grant Agreement No. 101143249 (HATEDEMICS)