AddressHate

AddressHate is a nonprofit working on detecting hate speech, focusing particularly on coded and implicit hate. Our technical team develops datasets, models, and tools to identify, annotate, and analyze hate online. We are currently focusing on antisemitism, anti-Black racism, and anti-LGBTQ hate.

Our work combines human annotation with multimodal machine learning, considering posts alongside their surrounding text and images.

What we build

• Annotation infrastructure for collecting and labeling antisemitism research data.
• Multimodal models for classification using text, images, and context.
• Analysis tools for exploring narratives, platforms, actors, and trends.

Resources

• AH Qwen3-VL-8B Antisemitism — a QLoRA adapter with PyTorch, Transformers, and PEFT usage examples.
• AH Antisemitism Multimodal — annotated text and images, with train and validation splits and a PyTorch DataLoader example.

These repositories are currently private and available to authorized collaborators.