Aspect-Based Sentiment Analysis with Dependency Relation Weighted Graph Attention

Aspect-based sentiment analysis is a fine-grained that focuses on the polarity of different aspects text, and most current research methods use combination dependent syntactic graphical neural networks. In this paper, graph attention network aspect-based model based weighting dependencies (WGAT) designed to address problem in traditional models do not sufficiently analyse types dependencies; proposed model, networks can be weighted averaged according importance nodes when aggregating information. The first transforms input text into low-dimensional word vector through pretraining, while generating dependency syntax by analysing constructing adjacency matrix graph. are then fed for feature extraction, predicted classification layer. focus more important during training, results comparison experiments Semeval-2014 laptop restaurant datasets ACL-14 Twitter social comment dataset show WGAT has significantly improved accuracy F1 values compared other baseline models, validating its effectiveness aspect-level tasks.