Multimodal Sentiment Analysis Using a Hybrid Deep Learning Model with Advanced Feature Fusion and Attention Mechanism

Multimodal sentiment analysis plays a crucial role in natural language processing by incorporating visual features, such as emojis, to interpret sentiment and mood in text. This paper addresses the challenge of effectively integrating textual and visual modalities, particularly the gap in leveraging emoji data for emotion detection. We propose a hybrid model that combines BERT-based textual embeddings with pretrained emoji embeddings, advanced feature fusion, and attention mechanisms to enhance sentiment understanding. The attention mechanism enables the model to focus on relevant semantic and emotional cues, while perceptual segmentation allows for improved interpretation of long-term dependencies between text and emojis. When evaluated on the Sentiment140 dataset, the proposed model achieves state-of-the-art performance, reaching an accuracy of 92.8%, Fl-score of 92.3%, and AUC of 96.2%, outperforming benchmarks like BERT-LSTM and RoBERTa-GRU. These results underscore the effectiveness of advanced feature fusion and attention mechanisms in multimodal sentiment analysis, providing robust emotion detection in social media data. Despite computational challenges, the model demonstrates strong generalization across different data types, offering valuable insights for future research in multimodal emotion analysis.