Improving Aspect-based Sentiment Analysis with Gated Graph Convolutional Networks and Syntax-based Regulation

Aspect-based Sentiment Analysis (ABSA) seeks to predict the sentiment\npolarity of a sentence toward a specific aspect. Recently, it has been shown\nthat dependency trees can be integrated into deep learning models to produce\nthe state-of-the-art performance for ABSA. However, these models tend to\ncompute the hidden/representation vectors without considering the aspect terms\nand fail to benefit from the overall contextual importance scores of the words\nthat can be obtained from the dependency tree for ABSA. In this work, we\npropose a novel graph-based deep learning model to overcome these two issues of\nthe prior work on ABSA. In our model, gate vectors are generated from the\nrepresentation vectors of the aspect terms to customize the hidden vectors of\nthe graph-based models toward the aspect terms. In addition, we propose a\nmechanism to obtain the importance scores for each word in the sentences based\non the dependency trees that are then injected into the model to improve the\nrepresentation vectors for ABSA. The proposed model achieves the\nstate-of-the-art performance on three benchmark datasets.\n