Graph Convolutional Network of Aspect-Level Sentiment Analysis Model with Sentiment Knowledge Enhancement

Aspect-level sentiment analysis is a meticulous task aimed at identifying the emotional tendencies of specific aspects or attributes in textual content. However, many existing related studies have overlooked the varying significance of different lexical components at a fine-grained level, resulting in models that are insensitive to syntactic structural information, underutilize external affective knowledge, and introduce extraneous information.To address these issues, we propose CSSGCN, a sentiment knowledge-enhanced GCN network model integrating semantic information with the syntactic structure of sentences.The proposed model employs the BERT pre-training model to acquire the initial word embeddings and enriches the features via CNN and a contextual position weight attenuation strategy. Next, a graph for GCN operations is constructed by integrating external sentiment knowledge with lexical constituent analysis. The model conducts a detailed analysis of lexical constituents such as negatives, adjectives, and verbs.Finally, both global and local information is retained to jointly achieve aspect-level sentiment classification.The validity of the model is verified through experiments on three publicly available datasets: Restaurant, Laptop, and Twitter, with accuracies of 85.49%, 80.92%, and 76.41% respectively, and corresponding F1 scores of 78.39%, 77.63%, and 75.51%.