Multimodal Fusion with Dual-Attention Based on Textual Double-Embedding Networks for Rumor Detection

Rumors may bring a negative impact on social life, and compared with pure textual rumors, online rumors multiple modalities at the same time are more likely to mislead users spread, so multimodal rumor detection cannot be ignored. Current methods for do not focus fusion of text picture-region object features, we propose neural network TDEDA (dual-attention based double embedding) applied detection, which performs high-level information interaction text–image level captures visual features associated keywords using an attention mechanism. In this way, explored ability enhance feature representation assistance from different in as well capture correlations dense between images text. We conducted comparative experiments two datasets. The experimental results showed that could reasonably handle thus improve accuracy currently relevant methods.