SentiVerse: A Transformer–BiLSTM-Based Explainable Framework for Cross-Modal Social Media Sentiment Analytics

Abstract

The exponential growth of social media platforms has generated vast amounts of heterogeneous data containing textual, visual, and contextual information, creating new opportunities for understanding human sentiments and emotions. However, conventional sentiment analysis approaches often focus on a single modality and lack transparency in their decision-making processes, limiting their effectiveness in dynamic and diverse online environments. This paper presents SentiVerse, an Explainable Cross-Modal Cognitive Intelligence Framework for dynamic sentiment and emotion analysis in social media. The proposed framework integrates textual, visual, and contextual features through a multimodal fusion architecture and employs a hybrid Transformer-BiLSTM network with attention mechanisms to capture semantic relationships and emotional patterns across different modalities. Explainable Artificial Intelligence (XAI) techniques based on SHAP and attention visualization are incorporated to provide interpretable predictions and enhance user trust. Experimental evaluation demonstrates that the proposed SentiVerse framework achieves 98.6% sentiment classification accuracy, 97.9% precision, 98.2% recall, and a 22.3% improvement in emotion recognition performance compared with conventional unimodal approaches. Furthermore, the framework effectively adapts to evolving social media trends and supports real-time cognitive analytics. The proposed system offers a robust and interpretable solution for intelligent sentiment understanding, opinion mining, and social media analytics applications.