Emotion-Enhanced Multi-Modal Persuasive Techniques Detection Using Split Features

The persuasive techniques in propaganda campaigns impact the Internet environment and our society. Detecting has aroused broad attention natural language processing field. In this paper, we propose a novel emotion-enhanced multi-level representation learning approach for multi-modal detection. To consider emotional factors used techniques, embed text images using different networks, use fully connected emotion enhanced layer to fuse embedding, where type strength of emotions are incorporated embedding. better model features fused inputted split-and-share module representations employed obtain detection performance. Furthermore, integrate focal loss alleviate problem data imbalance Experimental results on publicly dataset show that proposed is effective Remarkable experimental indicate capability MPDES extracting deeper information contained dual modalities.