<p>The Aspect Sentiment Triplet Extraction task involves identifying aspect terms, corresponding opinion terms, and their sentiment polarity within sentences, to comprehensively capture the text's fine-grained information. Existing methods predominantly utilize end-to-end approaches based on graph convolutional networks (GCNs). However, traditional GCN models may encounter long-distance dependency problems when dealing with longer sentences. Although increasing the number of GCN layers can help cover longer-distance dependencies, it leads to over-smoothing, where node features become indistinguishable due to the excessive aggregation of neighboring features as the number of layers increases. Therefore, in this article, we propose a GCN model enhanced by an edge attention transformer (EA-Transformer). We first extract local subgraph structures through GCNs and then capture global graph structures using the EA-Transformer method. Specifically, we use the EA-Transformer to learn different dependency types of adjacent edges; even edges with the same dependency type can obtain different representations and weights. In addition, when constructing edge representations, we incorporate multiple linguistic features and introduce an Additive Context Attention module, which enables the model to better capture and utilize critical contextual information. Extensive experiments on four benchmark datasets demonstrate that our model outperforms existing methods, achieving higher F1 scores while performing excellently in extracting overlapping triplets and handling long-distance dependencies.</p>