The main goal of the Aspect-based sentiment analysis (ABSA) task is to distinguish sentiment polarity in sentences based on aspect words. While previous methods, including attention mechanisms and graph-based approaches (GCNs), have made progress in capturing aspect-context relationships. These methods have difficulty in dealing with the changing relationship between the polarity and position of sentiment words in a sentence. Moreover, they often overlook inter-aspect sentiment dependencies within a sentence and without across sentences. We propose MoSGCN, a novel approach that addresses these limitations, on the one hand forming a distance matrix to adjust the distribution of sentiment words based on their proximity in the sentence, enhancing aspect-word relationships; the other a fine-tuned cross-similarity model based on Sentence Transformers to capture sentiment consistency across sentences. The results on multi-benchmark datasets reveal that MoSGCN surpasses prior models, proving its effectiveness in modeling intra- and inter-sentence sentiment relations. We have all codes and datasets at https://github.com/qxq-00/MoS-GCN .