Multi-refined Feature Enhanced Sentiment Analysis Using Contextual Instruction

TL;DR

CISEA-MRFE tackles domain-adaptive sentiment analysis by grounding deep PLM representations with contextual instructions, semantic augmentation, and multi-refined feature encoders. The framework integrates Contextual Instruction (CI), Semantic Enhancement Augmentation (SEA), Scale-Adaptive Depthwise Encoder (SADE) and Emotion Evaluator Context Encoder (EECE) to model local n-gram cues, global context, and affective signals. Across IMDb, Yelp, Twitter, and Amazon, it yields consistent accuracy and macro-F1 gains over strong baselines and remains effective across multiple PLMs, with competitive efficiency due to depth-wise convolutions. The work offers a scalable, generalizable approach to sentiment classification with practical deployment potential in diverse domains.

Abstract

Sentiment analysis using deep learning and pre-trained language models (PLMs) has gained significant traction due to their ability to capture rich contextual representations. However, existing approaches often underperform in scenarios involving nuanced emotional cues, domain shifts, and imbalanced sentiment distributions. We argue that these limitations stem from inadequate semantic grounding, poor generalization to diverse linguistic patterns, and biases toward dominant sentiment classes. To overcome these challenges, we propose CISEA-MRFE, a novel PLM-based framework integrating Contextual Instruction (CI), Semantic Enhancement Augmentation (SEA), and Multi-Refined Feature Extraction (MRFE). CI injects domain-aware directives to guide sentiment disambiguation; SEA improves robustness through sentiment-consistent paraphrastic augmentation; and MRFE combines a Scale-Adaptive Depthwise Encoder (SADE) for multi-scale feature specialization with an Emotion Evaluator Context Encoder (EECE) for affect-aware sequence modeling. Experimental results on four benchmark datasets demonstrate that CISEA-MRFE consistently outperforms strong baselines, achieving relative improvements in accuracy of up to 4.6% on IMDb, 6.5% on Yelp, 30.3% on Twitter, and 4.1% on Amazon. These results validate the effectiveness and generalization ability of our approach for sentiment classification across varied domains.

Figures (6)

• Figure 1: Overall framework of the proposed CISEA-MRFE model.
• Figure 2: Effect of input sequence length on classification accuracy across datasets. EECE benefits from longer sequences in datasets like IMDb, while SADE performs optimally at moderate lengths for shorter, noisier texts such as Twitter and Yelp.
• Figure 3: Comparison of feature fusion strategies between SADE and EECE outputs. Attention-based fusion yields superior performance across datasets by adaptively weighting local and global-emotional features.
• Figure 4: Performance of different language models in the text generation pipeline for data augmentation.
• Figure 5: Comparative performance analysis of various multi-kernel configurations in the SADE module across four benchmark datasets. The bar plot illustrates classification accuracy achieved by each kernel combination, while the overlaid line plot indicates the effective window size (maximum kernel span). The results indicate that integrating broader and more diverse receptive fields enhances local feature extraction, improving model performance across tasks.