Aspect-Based Sentiment Analysis (ABSA) is a fine-grained sentiment analysis task, aiming at mining sentiment polarity towards specific aspects. Most existing work to address ABSA has focused on using Graph Neural Networks combined with syntactic dependency trees. However, existing models often fall into semantic confusion for sentiment analysis due to the information imbalance in the dependency tree. To solve the problem of semantic confusion, we propose a Local Enhanced Relational Graph Attention Network with Dual-level Dependency Parsing (DL-...
