SENTICOG: CROSS-PLATFORM SENTIMENT ANALYSIS VIA CAUSAL GRAPH ENCODING AND MULTI-MODAL FUSION

In order to provide reliable and broadly applicable sentiment predictions, this study presents SENTICOG, a cross-platform sentiment analysis system that utilises Causal Graph Encoding and Multi-Modal Fusion. By merging textual features, graph-encoded causal linkages, and auxiliary metadata into a single latent representation, SENTICOG is able to capture deep semantic structures. Within the framework, two fundamental variations are investigated: CGR-MLP (Causal Graph Representation with Multi-Layer Perceptron) and CGR-ZSG (Causal Graph Representation with Zero-Shot Generalisation). CGR-ZSG achieves 89.6%, 93.23%, and 95.6% accuracy across benchmark datasets, allowing for generalisation to unknown domains without the need for retraining. CGR-MLP further improves classification performance with accuracies of 90%, 96.87%, and 97.78%, respectively, after being tuned for supervised tasks. These findings demonstrate the benefits of combining multi-modal learning and causal reasoning for scalable, cross-domain sentiment analysis.