Model Card for LS-W4-Aero-roberta-r4-hate-speech

Model Details

This is a 🤗 Transformers model card for a model on the Hugging Face Hub.

Model Description: LS-W4-Aero-roberta-r4-hate-speech is a hate speech detection model developed by Linkspreed UG. It's fine-tuned from the facebook/roberta-hate-speech-dynabench-r4-target model to identify hate speech in both English and German.

• Developed by: Linkspreed UG
• Shared by: Linkspreed UG
• Model type: Text Classification (Hate Speech Detection)
• Language(s) (NLP): English
• License: APACHE 2.0
• Finetuned from model: facebook/roberta-hate-speech-dynabench-r4-target

Uses

Direct Use

This model is for classifying text to detect hate speech, useful for content moderation, filtering, and analysis of online communications.

Out-of-Scope Use

This model should not be used for:

• Legal judgments or enforcing penalties.
• Automated censorship without human oversight.
• Applications infringing on free speech or human rights.
• Detecting subtle discrimination beyond direct hate speech.

Bias, Risks, and Limitations

Like all hate speech detection models, this model may show biases from its training data, potentially leading to:

• False Positives: Legitimate expressions misidentified as hate speech.
• False Negatives: Missed subtle or new forms of hate speech.
• Demographic Bias: Disproportionate flagging of content from specific groups if data was unbalanced.
• Language Nuance: Difficulty with contextual or evolving hate speech.

Recommendations

Users should be aware of the model's risks, biases, and limitations. Human review is strongly recommended for critical applications. Regular auditing of performance on diverse datasets is advised.