This research introduces a keyword-oriented multimodal approach to identify euphemisms in online content, crucial for security and content moderation applications.

• Combines text, image, and audio analysis to detect disguised references to illicit content
• Addresses the growing challenge of euphemism use in underground markets (drugs, weapons, explicit content)
• Overcomes the limitation of text-only approaches that miss contextual visual and audio cues
• Provides a foundation for more effective content moderation systems in multimodal social media environments

This work offers significant security applications by enabling platforms to better identify and moderate potentially harmful content disguised through euphemistic language across multiple media formats.

Keyword-Oriented Multimodal Modeling for Euphemism Identification