Improved Adjacency Matrix-Based Graph Convolutional Network Aspect-Level Sentiment Analysis
Lìjiāng Liú, Hongying Lu, Yao Wang, Haifeng Li · 2022 5th International Conference on Pattern Recognition and Artificial Intelligence (PRAI) · 2022
Existing aspect-based sentiment classification methods use graph-based models to integrate the syntactic structure of sentences. Although these methods are practical, they give less consideration to the aspect-context relative position distance relationship when creating the adjacency matrix, and ignore the point that contexts with different distances from aspects have different effects on the feature extraction of aspects, and simply give the same weight to contexts only. To address the above problems, this paper proposes to fuse the context and aspect relative position distance weight relationship in the syntactic relationship adjacency matrix of sentences. Firstly, the context information of the sentence is obtained using a two-way long and short-term memory network. Then the adjacency matrix is created based on the aspect-context relative distance and syntactic relationship. Finally, a graph convolutional network is used to extract the sentiment features of aspects by combining sentence context and syntactic relations. The experimental results on five public datasets demonstrate that the adjacency matrix incorporating the distance relationship is effective.