Toward a Dual Attention Model for Image-Text Sentiment Classification
Soukaina Fatimi, Wafae Sabbar, Abdelkrim Bekkhoucha · 2023
Users are becoming accustomed to uploading text and pictures on social networks to express their feelings or ideas. As a result, multimodal sentiment analysis has drawn more attention as a study area in recent years. Nevertheless, most of the existing multimodal sentiment analysis approaches are focused either on emphasising the sentiment features in textual-visual data or on giving more importance to the correlation between texts and images. Motivated by this observation, we propose a dual attention model for image-text sentiment classification that focuses on both extracting attended visual and textual features relevant to the semantic classification, and uses the selected features to compute the correlation between the visual and textual data. Moreover, we take advantage of both the self-attention mechanism and the cross-modal mechanism to put in place a dual attention model for multimodal sentiment classification. The approach is carefully constructed and extensively explained, offering a strong theoretical base. The current research focuses on presenting a new theoretical method and its overall architecture. However, experimental validation is a crucial next step; a later article will present the experimental findings.