To enhance the accuracy and efficiency of aspect sentiment triplet extraction (ASTE), this study proposes an ASTE model that integrates graph convolution networks with graph attention networks. The model incorporates graph convolution networks within a bidirectional machine reading comprehension framework and further integrates a multi-head graph attention mechanism. By leveraging self-attention, dynamic weight allocation is achieved, enabling more effective identification of complex and implicit semantic relationships and more precise extraction of sentiment expressions. In addition, contrastive learning is employed to enhance global semantic representations, and reinforcement learning is introduced to refine the extraction strategy through agent-environment interaction and reward-guided optimization. Extensive experiments conducted on four public datasets demonstrate that the proposed model achieves superior performance on the ASTE task, with significant improvements in accuracy, recall, and F1 compared to existing approaches.