Can Thinking Models Think to Detect Hateful Memes?

2026-03-01

AI Summary

Detecting hateful memes requires precise understanding of the implicit harmful intent conveyed through the synergistic interplay of text and image, posing a significant challenge for fine-grained multimodal reasoning. This work proposes a reinforcement learning–based post-training framework that jointly optimizes both classification performance and explanation generation quality under weak supervision, integrating chain-of-thought distillation with a newly designed Group Relative Policy Optimization (GRPO) objective. Evaluated on the Hateful Memes benchmark, the method achieves state-of-the-art results, improving accuracy and F1 score by approximately 1% and enhancing explanation quality by about 3%. These gains substantially advance the interpretability and robustness of multimodal large language models in detecting hate content.

Technology Category

• Machine Learning: Multimodal Learning
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Hateful memes often require compositional multimodal reasoning: the image and text may appear benign in isolation, yet their interaction conveys harmful intent. Although thinking-based multimodal large language models (MLLMs) have recently advanced vision-language understanding, their capabilities remain underexplored for hateful meme analysis. We propose a reinforcement learning based post-training framework that improves reasoning in thinking-based MLLMs through task-specific rewards and a novel Group Relative Policy Optimization (GRPO) objective. Specifically, we (i) conduct a systematic empirical study of off-the-shelf MLLMs for hateful meme understanding, (ii) extend an existing hateful meme dataset by generating weakly or pseudo-supervised chain-of-thought rationales via distillation, and (iii) introduce a GRPO-based objective that jointly optimizes meme classification and explanation quality to encourage fine-grained, step-by-step reasoning. Experiments on the Hateful Memes benchmark show that our approach achieves state-of-the-art performance, improving accuracy and F1 by approximately 1 percent and explanation quality by approximately 3 percent. We will publicly release our code, dataset extensions, and evaluation resources to support reproducibility.

Problem

• hateful memes
• multimodal reasoning
• multimodal large language models
• compositional reasoning
• vision-language understanding

Innovation

• multimodal reasoning
• reinforcement learning
• chain-of-thought
• Group Relative Policy Optimization
• hateful meme detection