Multi-hop Syntactic Graph Convolutional Networks for Aspect-Based Sentiment Classification

Sentiment analysis is widely applied to online and offline applications such as marketing, customer service and social media. Aspect-based sentiment classification is a fine-grained sentiment analysis that identifies the sentiment polarity of a specific aspect in a given sentence. In order to model syntactical constraints and word dependencies in a sentence, graph convolutional network (GCN) has been introduced for aspect-based sentiment classification. Though achieved promising results, GCN becomes less effective when the aspect term is far from the key context words on the dependency tree. To tackle this problem, we propose a Multi-hop Syntactic Graph Convolutional Networks model, in which a syntactic graph convolutional network is constructed according to transmission way of information in the sentence structure. Then a multi-range attention mechanism is applied to deepen the number of layers of the model to aggregate further information on the dependency tree. Experiments on benchmarking collections show that our proposed model outperforms the state-of-the-art methods.