hARmful Meme

Harmful memes, combining text and images to spread harmful ideologies or attack individuals, are a growing concern requiring automated detection. Current research focuses on developing multimodal models, often leveraging large language models (LLMs) and incorporating techniques like image captioning and optical character recognition (OCR), to analyze both visual and textual content for nuanced understanding of toxicity. This work is crucial for improving online safety and informing the development of effective content moderation strategies, with a particular emphasis on explainability and addressing challenges posed by multilingual and low-resource contexts.

Papers