Answers are grounded in this paper and cite their sources.
Review highlights challenges in hate speech detection and multimodal classification methods on social media.
With the explosive growth of social media content and the enhanced information-sharing capabilities of platforms, the proliferation of online hate speech has become a global governance challenge. Its dissemination patterns are rapidly evolving towards multimodality (the deep integration of text and images), further complicating content security management. On one hand, the subtlety and strong contextual dependence of hate speech significantly increase the difficulty of detection. On the other hand, emerging forms of dissemination, such as meme images, present dual challenges for classification tasks due to their inherent characteristics: a seemingly humorous facade, reliance on cultural context, and semantic conflicts between text and image. To address these issues, this paper focuses on two major technical approaches: unimodal text analysis and multimodal content classification. It provides a systematic review of the research progress in hate speech detection methods based on text and multimodal detection methods, while also analyzing their limitations. Furthermore, this paper consolidates the characteristics and applicable scenarios of current mainstream unimodal and multimodal hate speech datasets, offering reference directions for optimizing technical approaches and constructing datasets in subsequent research.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: