Multi-Channel Relational GCN Adaptive Fusion for Multi-Aspect Sentiment Classification

Most existing Aspect-Based Sentiment Analysis (ABSA) models predict the sentiment polarity of only one aspect at a time, focusing on enhancing the representation of this single aspect based on other contexts or aspects. This one-to-one paradigm neglects that multi-aspect, multi-sentiment sentences not only contain specific descriptions of different particular aspects but also share global contextual information across multiple aspects. Recent methods have achieved simultaneous prediction of multiple aspect sentiments by using multiple relational graph neural networks to comprehensively capture the interactions between different aspects within a sentence and between aspects and context, improving detection efficiency. We argue that different relationships are of varying importance for aspect sentiment analysis. Therefore, we propose an adaptive weighting method for multi-channel relational graph convolutional networks. Extensive experiments were conducted on three public datasets (MAMS, Rest14, and Lap14). The results demonstrate the effectiveness of our method in multi-aspect, multi-sentiment scenarios.