Training-Free and Interpretable Hateful Video Detection via Multi-stage Adversarial Reasoning

TL;DR

The paper tackles hateful video content detection under data scarcity and interpretability constraints. It introduces MARS, a training-free multi-stage adversarial reasoning framework that uses prompting of visual-language models to generate an objective content description, followed by hate and non-hate hypotheses and a meta-analytical synthesis to produce a final decision. MARS provides explicit, evidence-based explanations, reduces false positives, and shows competitive performance against training-based models on real-world datasets HateMM and MHC, with demonstrated multilingual robustness. The work advances auditable content moderation by delivering transparent justifications and releasing code for reproducibility.

Abstract

Hateful videos pose serious risks by amplifying discrimination, inciting violence, and undermining online safety. Existing training-based hateful video detection methods are constrained by limited training data and lack of interpretability, while directly prompting large vision-language models often struggle to deliver reliable hate detection. To address these challenges, this paper introduces MARS, a training-free Multi-stage Adversarial ReaSoning framework that enables reliable and interpretable hateful content detection. MARS begins with the objective description of video content, establishing a neutral foundation for subsequent analysis. Building on this, it develops evidence-based reasoning that supports potential hateful interpretations, while in parallel incorporating counter-evidence reasoning to capture plausible non-hateful perspectives. Finally, these perspectives are synthesized into a conclusive and explainable decision. Extensive evaluation on two real-world datasets shows that MARS achieves up to 10% improvement under certain backbones and settings compared to other training-free approaches and outperforms state-of-the-art training-based methods on one dataset. In addition, MARS produces human-understandable justifications, thereby supporting compliance oversight and enhancing the transparency of content moderation workflows. The code is available at https://github.com/Multimodal-Intelligence-Lab-MIL/MARS.

Figures (2)

• Figure 1: Overview of the proposed multi-stage adversarial reasoning framework.
• Figure 2: Examples of interpretability. Red lines indicate corresponding factors and video content.