Aspect-based sentiment analysis aims to detect the sentiment polarity of specific aspects in text, playing a critical role in product review analysis and public opinion monitoring. Although pre-trained language models have significantly advanced sentiment analysis, they still suffer from limitations in deep semantic-syntactic fusion and cross-aspect sentiment relation modeling. To address these gaps, this paper proposes a BERT-based framework integrated with a semantic-syntactic dual-channel mechanism. The semantic channel enriches aspect representations by fusing a sentiment lexicon with component parse trees, further refined via hierarchical graph learning; the syntactic channel leverages dependency parsing graphs to enhance structured semantic inference. Additionally, an attention-guided lowest common ancestor algorithm is introduced to capture cooperative and contrastive relations between different aspects. Experimental results demonstrate that the deep fusion of dual channels outperforms single-channel variants, and ablation studies show significant performance degradation when disabling the interaction module, validating the model’s effectiveness in leveraging cross-aspect emotional cues.