Aspect term extraction is a crucial step in aspect-level sentiment analysis, significantly affecting the accuracy of sentiment classification. Therefore, improving the precision of aspect term extraction is vital for enhancing the performance of sentiment analysis. The limitations of existing methods include inadequate consideration of syntactic information and inter-word dependencies, as well as the challenge of mitigating weight noise during dependency tree conversion. To address these issues, we propose an aspect term extraction approach that leverages dynamic attention and graph convolutional network. Our method utilizes a densely connected graph convolutional network to capture dependency information between distant terms, thereby enriching vector semantics. Furthermore, it integrates a dynamic attention mechanism informed by dependency parsing to highlight critical dependencies and mitigate noise interference. We benchmark our model against state-of-the-art approaches on four widely used public datasets. The results indicate that our proposed method significantly enhances the performance of aspect term extraction. Specifically, our model improves upon baseline models on the Lap14 and Rest15 datasets, with increases in macro-F1 scores of 0.45, and 0.04, respectively.