New graph networks enhance detection of multimodal sarcasm and cyberbullying

Researchers have developed two novel frameworks, HCIG and GCCN, designed to improve the detection of multimodal sarcasm and cyberbullying. HCIG, the Hierarchical Cross-modal Incongruity Graph Network, models incongruity across token, phrase, and global levels using graph attention networks. GCCN, the Graph-based Cross-modal Contradiction Network, employs graph-based reasoning with contradiction-aware pooling. Both models aim to capture semantic inconsistencies between text and visual information more effectively than traditional fusion methods. Experiments on benchmark datasets showed HCIG achieving high accuracy on sarcasm detection and GCCN excelling in cyberbullying detection. AI

IMPACT These graph-based methods offer a more nuanced approach to understanding multimodal content, potentially improving AI's ability to interpret complex social interactions online.

RANK_REASON The cluster describes a new academic paper proposing novel methods for a specific AI task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Dinesh Kumar Vishwakarma
• Graph-based Cross-modal Contradiction Network
• HCIG
• Hierarchical Cross-modal Incongruity Graph Network
• MultiBully

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →