Local Dependency-Enhanced Graph Convolutional Network for Aspect-Based Sentiment Analysis

The task of aspect-based sentiment analysis (ABSA) is to detect the polarity toward given aspects. Contemporary methods predominantly utilize graph neural networks and incorporate attention mechanisms dynamically connect aspect terms with their surrounding contexts, resulting in more informative feature representations. However, these only consider whether there are dependencies between words when introducing dependencies, ignoring that different have effects. Neglecting this could introduce noise negatively impact model’s performance. To overcome limitation, we a novel approach called local dependency-enhanced convolutional network (LDEGCN). Our method combines semantic information dependency relationships better capture affective words. Specifically, integrate knowledge from SenticNet enrich sentence’s thoroughly explore types contexts aspects focus on particular types. context weight (LCW) employed emphasize importance thereby mitigating issue long-distance dependencies. Through extensive evaluations five public datasets, LDEGCN model demonstrates significant improvements over mainstream models.