<p>Aspect sentiment triplet extraction aims to extract aspect terms, opinion terms, and sentiment polarity as triplets. Recent studies tend to approach this task via table-filling methods, often in conjunction with dependency tree-based graph Convolution Neural Network. However, these methods determine sentiment polarity solely through downstream region classification processes, overlooking the sentiment tendencies embedded in the original text. As a result, they offer limited guidance from the original text for identifying sentiment polarity, leading to weak connections between sentiment and aspect-opinion pairs. To tackle this issue, we integrate prompt representations and context representations through a novel transformer. Specifically, we employ a prompt template to extract sentiment representations during sentence encoding. Afterward, the sentence representation undergoes a process of dependency graph construction and graph convolution to extract contextual feature representations. Subsequently, the contextual feature representations and prompt representations are interactively integrated through Cross-Attention Diffusion Transformer, and triplets are obtained using a table-filling method. During the construction of the dependency graph, potential aspect terms and opinion terms are predicted based on part-of-speech to enhance the dependency graph. In addition, our proposed Cross-Attention Diffusion Transformer facilitates better alignment between sentiment information extracted from the original text and the regional classification process. Moreover, considering the inconsistent accuracy in identifying aspect and opinion terms based on part-of-speech, we design a label update module that only updates the aspect prediction labels within the search domain while retaining the original opinion prediction labels. Experiments on ASTE-Data-V2 dataset demonstrate that our method achieves state-of-the-art performance.</p>