Multimodal Aspect-Level Sentiment Analysis Based on Attention Mechanism and Multi-Layer Perceptron

Multimodal Aspect-Based Sentiment Analysis aims to enhance sentiment classification accuracy by integrating information from multiple modalities, such as text and images. This paper introduces an enhanced method for multimodal sentiment analysis, leveraging the FITE model along with Multi-Layer Perceptron (MLP) and Multi-Head Self-Attention for better feature fusion. The approach first designs an effective fusion strategy to improve the interaction between modalities, then applies a gating mechanism to prioritize textual and visual information. Experiments on Twitter-2015 and Twitter-2017 datasets show the model outperforms existing baselines in precision and F1 score. The model also demonstrates robustness and adapts well to modality variations. Future work will aim at improving cross-domain adaptability, optimizing feature fusion, and applying the model in real-time sentiment analysis.