Dynamic position weighting aspect-focused graph convolutional network for aspect-based sentiment analysis

Aspect-based sentiment analysis (ABSA) is a fine-grained sentiment analysis task that analyzes the affective attitudes of specific aspects of a review. Recent studies have focused on using graph convolutional networks and attention mechanisms for ABSA; however, most of the existing works fail to flexibly consider the internal distance relationships between aspects and contexts when constructing dependency graphs, and their models do not pay sufficient attention to the aspects in the feature extraction process after performing graph convolution. In this paper, we propose a dynamic position weighting aspect-focused graph convolutional network (DPWAFGCN-BERT) to address the above problems. Specifically, we combine the relative distance and dependency distance measures to weight the original dependency graph and utilize dynamic coefficients to control the influence strengths of different distance types to achieve enhanced aspect sentiment feature aggregation. Furthermore, after implementing graph convolution, we design an aspect-focused attention fusion module, which includes both a retrieval-based multihead attention mechanism and an aspect-oriented multihead attention mechanism, to learn contextual sentiment features based on aspects from different feature subspaces. We conduct experiments on four public datasets, and the experimental results demonstrate the excellent performance of our proposed model.