Long-distance sentiment dependency strengthened graph convolutional network for aspect-based sentiment analysis

As a detailed sentiment categorization task, Aspect-based Sentiment Analysis (ABSA) focuses on accurately identifying the emotional polarity of concrete aspect items within a sentence. Researchers have made significant advancements by integrating sentence dependency structures with Graph Neural Networks (GNN) over recent years. However, existing methods still exhibit limitations in extracting potential information from syntactic structures; they often overlook words that are critical for assessing aspect terms and demonstrate constrained performance in capturing contextual semantic information. We introduce an innovative model to overcome such challenges: the Long-Distance Sentiment Dependency Enhanced Graph Convolutional Network (LDSDGCN). This model transforms the traditional sequential context extraction approach into a structured context extraction method through the introduction of Tree-LSTM. Additionally, it constructs a long-distance emotional dependency-strengthened graph convolutional network to pick up higher-order syntax information among words. Through experimental validation on three communal datasets, our pattern significantly surpasses existing baseline methods, thereby showcasing its efficiency in obtaining long-distance dependencies and complicated syntactic relationships.