SyRHM Framework for Zero-shot Harmful Meme Detection
September 15, 2026
SyRHM detects harmful memes by decomposing multimodal content into textual descriptions and using symbolic-language-enhanced reasoning. The framework uses a planner-solver architecture to interpret visual-textual incongruity and cultural stereotypes.
HOW THIS AFFECTS YOU
●
researcherThe multi-stage symbolic reasoning approach offers a more interpretable path for multimodal safety detection.
●
policyThis provides a technical mechanism for more nuanced content moderation of implicit harmful intent.