This paper describes our system used in the SemEval-2026 Task 3: Dimensional Aspect-Based Sentiment Analysis. To address the inherent subjectivity and nuanced emotional expressions in this task, we propose a Retrieval-Augmented Generation (RAG) framework based on Large Language Models (LLMs) for sentiment triplet extraction. Our approach leverages a dynamic retrieval mechanism to identify semantically similar training examples, which are then integrated into the prompts as in-context demonstrations. This strategy effectively guides the model’s inference process by providing relevant linguistic patterns and emotional contexts. Our implementation is available at https://github.com/Kevinyu66/dimaste.