Aspect-level sentiment analysis based on window-aware syntactic and semantic graph convolutional network

With the widespread application of aspect-level sentiment analysis, graph convolutional networks integrating semantic and syntactic information have significantly enhanced sentiment classification performance. However, while traditional graph convolutional networks effectively capture local dependencies, they often introduce irrelevant contexts and redundant connections, compromising the accuracy of graph structure learning. To address this issue, this paper proposes an aspect-level sentiment analysis method based on window-aware syntax and semantic graph convolutional networks. The approach incorporates a sliding window strategy within the attention mechanism to enhance local context perception and improve long-distance dependency modeling. Additionally, it integrates dilated convolution and multi-layer residual connections within the graph convolutional network to expand the receptive field, enhance information propagation efficiency, and mitigate gradient vanishing. Furthermore, a cross-layer residual connection is introduced in the information aggregation stage to establish a dynamic interaction mechanism across multiple feature layers, enhancing the complementarity of semantic and syntactic information. Experimental results on four public datasets demonstrate that the proposed model outperforms existing benchmark methods.