VISUALIZING MISOGYNY: A CORPUS ASSISTED MULTIMODAL ANALYSIS OF ONLINE ABUSE AND DENIGRATION

Abstract

Internet memes, as an increasingly prevalent medium of online communication, often carry hidden ideological content. This study investigates how misogynistic ideologies are encoded and disseminated through these memes, addressing a research gap across feminist media studies and computational discourse analysis by focusing on Reddit memes from Western and Pakistani social media contexts. Drawing on Feminist Critical Discourse Analysis (FCDA) and intersectionality theory, the research employs a corpus-assisted multimodal discourse analysis framework to examine gendered content in memes. A dataset of roughly 1,800 memes is compiled and classified via optical character recognition (OCR) and large language models into thematic categories such as shaming, stereotyping, objectification, and violence. The methodology integrates natural language processing for textual analysis with LLM‘s vision capabilities for visual content analysis, thereby capturing both verbal and visual elements. Multimodal classification uncovers how humor and visual rhetoric mask misogynistic messages, facilitating their normalization. Key findings reveal that memes often leverage humor, irony, and culturally specific imagery to disguise abuse and ridicule, yet underlying themes of gendered denigration remain consistent across samples. Cross-cultural comparison identifies both global consistencies in sexist memes and local specificities shaped by cultural norms and x intersectional identities. By highlighting these patterns of encoded misogyny in everyday digital content, the study elucidates how online humor and imagery perpetuate sexist ideologies. This research contributes to feminist media studies, digital discourse analysis, and automated misogyny detection by illuminating covert mechanisms of online gendered abuse and proposing a methodology for analyzing multimodal hate content in digital media discourse.