Aspect-based sentiment analysis (ABSA) focuses on identifying sentiment polarity toward specific aspects in text and is critical for fine-grained opinion understanding. Existing graph-based methods often fail to capture long-range sentiment dependencies and struggle to distinguish sentiment-aspect associations in sentences with multiple sentiment expressions. To address these issues, we propose SARGAT (Sentiment-Aspect Relational Graph Attention Network), a novel model that jointly enhances structural representation and relation modeling. First, we introduce a sap-RGAT module that extends the dependency graph with a new edge type, sap, explicitly connecting aspect words to sentiment words. This enhancement preserves long-distance sentiment flows and mitigates performance degradation in lengthy or syntactically complex inputs. Second, we design SD-Attention, a sentiment distance-aware attention mechanism that integrates dependency and relative positional distances into a unified representation, enabling finer-grained modeling of sentiment-aspect interactions. Experiments on multiple benchmark ABSA datasets demonstrate that SARGAT outperforms strong baselines, with average improvements of 1.93% in accuracy and 1. 5% in F1 score. The results confirm that SARGAT effectively models complex sentiment structures, offering a lightweight and interpretable solution for aspect-level sentiment analysis.