Synthetic Sentiment Cue Enhanced Graph Relation-Attention Network for Aspect-Level Sentiment Analysis

Aspect-level Sentiment Analysis (ASA) constitutes a critical task in natural language processing, aimed at detecting and classifying sentiment polarities associated with specific aspects in textual data. While conventional approaches often fall short in capturing fine-grained sentiment distinctions at the aspect level, neural network-based models, despite their advancements, frequently encounter challenges stemming from insufficient explicit emotional indicators such as aspect-sentiment correlations, contextual dependencies, and salient opinion expressions. To address these limitations, this paper presents a novel Synthetic Sentiment Cue Enhanced Graph Relation-Attention Network (SSC-GRAN), a hybrid framework that synergistically integrates large language models (LLMs) with graph neural networks (GNNs). Our method introduces three key innovations: (1) a synthetic data augmentation paradigm leveraging LLMs to generate semantically coherent sentiment cues, thereby enriching aspect-opinion interactions; (2) a hierarchical graph architecture that models syntactic dependency structures and aspect-context relationships through relation-aware attention mechanisms; and (3) a contrastive learning objective that aligns representations from both authentic and synthetic data to enhance model robustness. Comprehensive experiments across benchmark datasets demonstrate statistically significant improvements in performance, with SSC-GRAN achieving state-of-the-art results. Ablation studies further validate the contributions of each component, while qualitative analyses reveal enhanced interpretability in aspect-sentiment reasoning. This work advances ASA research by bridging the gap between data scarcity and structural modeling, offering a robust framework for fine-grained sentiment understanding.