DyFuLM: An Advanced Multimodal Framework for Sentiment Analysis

TL;DR

This work tackles the challenge of fine-grained sentiment analysis in complex text by introducing DyFuLM, a multimodal framework that jointly models coarse sentiment, fine-grained emotions, and emotion intensity through hierarchical dynamic fusion and gated feature aggregation. By leveraging a dual-encoder backbone (RoBERTa and DeBERTa) and a BiLSTM-guided cross-layer fusion, the model adaptively integrates multi-level semantics and cross-model cues. Extensive experiments on a large hotel-review dataset demonstrate superior performance across classification and regression tasks, with ablation studies confirming the contribution of each module. The approach offers robust multi-dimensional sentiment representations with potential for domain adaptability and improved interpretability in affective computing applications.

Abstract

Understanding sentiment in complex textual expressions remains a fundamental challenge in affective computing. To address this, we propose a Dynamic Fusion Learning Model (DyFuLM), a multimodal framework designed to capture both hierarchical semantic representations and fine-grained emotional nuances. DyFuLM introduces two key moodules: a Hierarchical Dynamic Fusion module that adaptively integrates multi-level features, and a Gated Feature Aggregation module that regulates cross-layer information ffow to achieve balanced representation learning. Comprehensive experiments on multi-task sentiment datasets demonstrate that DyFuLM achieves 82.64% coarse-grained and 68.48% fine-grained accuracy, yielding the lowest regression errors (MAE = 0.0674, MSE = 0.0082) and the highest R^2 coefficient of determination (R^2= 0.6903). Furthermore, the ablation study validates the effectiveness of each module in DyFuLM. When all modules are removed, the accuracy drops by 0.91% for coarse-grained and 0.68% for fine-grained tasks. Keeping only the gated fusion module causes decreases of 0.75% and 0.55%, while removing the dynamic loss mechanism results in drops of 0.78% and 0.26% for coarse-grained and fine-grained sentiment classification, respectively. These results demonstrate that each module contributes significantly to feature interaction and task balance. Overall, the experimental findings further validate that DyFuLM enhances sentiment representation and overall performance through effective hierarchical feature fusion.

Figures (6)

• Figure 1: Flowchart of Data Preprocessing
• Figure 2: DyFuLM Workflow
• Figure 3: Comparison between BERT and DyFuLM architectures (green denotes shared components, red denotes DyFuLM-specific modules)
• Figure 4: Emotion Category Distribution of Comments
• Figure 5: Percentage of Performance Improvement