Sentiment analysis using multi neighborhood representation learning embeddings: graph convolutional neural networks

Abstract

With the thriving of social networks, sentiment analysis at aspect-level has become a research hotspot in the field of natural language processing, aiming to extract user expressed sentiment from massive text data. To tackle the problem that current sentiment analysis methods cannot simultaneously consider the contextual and syntactic information of aspect terms, this paper proposes a targeted model for aspect-level sentiment analysis by embedding different neighborhood representations. By integrating aspect neighborhood information and node neighborhood information through convolutional neural networks, our proposed model achieves high effectiveness on five public datasets as demonstrated by the evaluation coefficients constructed in our study.