思考模型能否通过思考来检测仇恨言论？
Can Thinking Models Think to Detect Hateful Memes?
AI总结：
本文提出基于强化学习的后训练框架，通过任务特定奖励和GRPO目标提升基于思考的多模态模型在检测仇恨言论中的推理能力，实验表明在准确性、F1值和解释质量上均有显著提升。
摘要：
仇恨言论往往需要组合性的多模态推理：图像和文本在孤立状态下可能显得无害，但它们的相互作用传达了有害的意图。尽管基于思考的多模态大语言模型（MLLMs）最近在视觉-语言理解方面有所进展，但它们在仇恨言论分析中的能力仍然未被充分探索。我们提出了一种基于强化学习的后训练框架，通过任务特定的奖励和一种新的组相对策略优化（GRPO）目标来改进基于思考的MLLMs的推理能力。具体来说，我们（i）对现成的MLLMs进行了系统性的实证研究，以进行仇恨言论理解；（ii）通过蒸馏生成弱监督或伪监督的推理链条；（iii）引入了一种基于GRPO的目标，共同优化膜分类和解释质量，以鼓励细粒度、逐步推理。在仇恨言论基准测试上的实验表明，我们的方法实现了最先进的性能，将准确率和F1值提高了约1个百分点，将解释质量提高了约3个百分点。我们将公开发布我们的代码、数据集扩展和评估资源以支持可重复性。
英文摘要：
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.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
