FBHM: Functional Benchmarking and Steering of VLMs for Hateful Meme Detection

Hateful meme detection remains a formidable challenge for vision-language models, as existing benchmarks are structurally observational - confounding rhetorical hate mechanisms with target community features and preventing causal evaluation of model vulnerabilities. To address this, we introduce FBHM, a systematically curated benchmark of Functionality Based Hateful Memes constructed along two orthogonal axes: 25 distinct rhetorical functionalities and 10 target communities (5,000 memes total). Benchmarking state-of-the-art VLMs reveals a severe generalization gap: models highly accurate on standard datasets catastrophically drop to near-random performance on FBHM, proving they exploit dataset-specific heuristics rather than robust multimodal reasoning. To efficiently close this gap, we propose LSV (learnable steering vectors), an ultra-low data regime strategy that applies a causal intervention objective on as few as 500 steering samples (50 unique base memes), boosting FBHM performance by ~30 Macro-F1 points while outperforming in-context learning and PEFT without degrading source-domain performance.

See, Explain, and Intervene: A Few-Shot Multimodal Agent Framework for Hateful Meme Moderation

In this work, we examine hateful memes from three complementary angles - how to detect them, how to explain their content and how to intervene them prior to being posted - by applying a range of strategies built on top of generative AI models. To the best of our knowledge, explanation and intervention have typically been studied separately from detection, which does not reflect real-world conditions. Further, since curating large annotated datasets for meme moderation is prohibitively expensive, we propose a novel framework that leverages task-specific generative multimodal agents and the few-shot adaptability of large multimodal models to cater to different types of memes. We believe this is the first work focused on generalizable hateful meme moderation under limited data conditions, and has strong potential for deployment in real-world production scenarios. Warning: Contains potentially toxic contents.

Zero shot VLMs for hate meme detection: Are we there yet?

Multimedia content on social media is rapidly evolving, with memes gaining prominence as a distinctive form. Unfortunately, some malicious users exploit memes to target individuals or vulnerable communities, making it imperative to identify and address such instances of hateful memes. Extensive research has been conducted to address this issue by developing hate meme detection models. However, a notable limitation of traditional machine/deep learning models is the requirement for labeled datasets for accurate classification. Recently, the research community has witnessed the emergence of several visual language models that have exhibited outstanding performance across various tasks. In this study, we aim to investigate the efficacy of these visual language models in handling intricate tasks such as hate meme detection. We use various prompt settings to focus on zero-shot classification of hateful/harmful memes. Through our analysis, we observe that large VLMs are still vulnerable for zero-shot hate meme detection.