Aspect-level Sentiment Classification with Multi-head-attention-based Multi-channel Graph Convolutional Networks

Abstract

Aspect-level sentiment classification (ALSC) has received extensive attention due to its fine-grained characteristics. However, ALSC tasks are often plagued by complex sentence structures, and existing attention-based methods cannot fully utilize implicit syntactic information to deconstruct sentences. Therefore, we propose a model named MAMC-GCN (Multi-head-attention-based Multi-channel Graph Convolutional Networks), which integrates the dependency relations of sentences and uses the multi-head attention mechanism to improve the ability to extract key information related to aspect terms. Experiments on five benchmarks prove that our model can make excellent performance under a short training duration.