Aspect-level sentiment analysis, as a sentiment analysis task, aims to identify the sentiment toward specific aspects or topics mentioned in text. To optimize its performance, which is constrained by internal text information and ignores features such as part-of-speech, dependency relationship types, and syntactic distance in the syntactic dependency graph that could enhance the semantic information of aspect words, this paper combines the syntactic dependency graph with a graph neural network model. By leveraging external knowledge to enhance the graph attention network, we propose a graph neural network-based aspect-level text sentiment analysis method focused on semantic enhancement. We collect theme-related comment text data from various social media platforms to create a dataset tailored for aspect-level sentiment analysis composite tasks. By comparing with multiple baseline methods, we analyze the advantages of the proposed model in Chinese semantic analysis applications. The proposed model, SEGCN, achieves semantic analysis accuracy rates of 92.23% (PTS-1), 83.45% (PTS-2), and 90.76% (PTS-3) across all datasets, outperforming other baseline methods.
