Aspect-based sentiment analysis (ABSA) is a natural language processing (NLP) task that identifies sentiment polarity toward specific aspects in a sentence, which is crucial for fine-grained opinion mining. Recent advances in deep learning have leveraged data augmentation and contrastive learning to improve sentence representations for ABSA, enabling models to distinguish between sentiment polarities. However, existing augmentation methods often rely on rigid rules, failing to capture nuanced contextual relationships. To overcome this limitation, we propose using large language models (LLMs) to generate context-aware augmented samples tailored to target datasets. Furthermore, prior work has largely ignored that the same aspect term can express divergent sentiments across contexts. We address this by introducing a Dual Contrastive Learning for Aspect and Sentence (DCAS) model, which jointly optimizes aspect and sentence level representations. Experiments on four benchmark datasets validate our approach’s effectiveness.