A survey of multimodal hate meme detection

Memes have emerged as a powerful tool for overt hate speech as rapidly growing social media platforms respond to each other. They use metaphors and subtle cues rooted in cultural contexts, making automatic detection extremely challenging. Although previous research has primarily focused on improving multimodal fusion methods, relatively few studies have explicitly examined how cross-modal interaction shapes the semantic processes underlying implicit hatred. These gaps are the ones this survey will endeavor to fill, though it will also offer an overview of the most recent developments in hateful meme detection. It examines multimodal representations in light of the underlying semantic processes that express implicit hate. Recent methods of detection could be divided into two major groups (fusion-based and fusion-free). Popular datasets across various settings are also examined in the survey, major challenges are identified, and future research directions are suggested, which can be useful to researchers in the field