DAGCN: Distance-based and Aspect-oriented Graph Convolutional Network for Aspect-based Sentiment Analysis

Zhihao Wang, Bo Zhang, Ru Yang, Chang Guo, Maozhen Li

Primary Area: representation learning for computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Aspect-based sentiment analysis, Graph convolutional network, Kullback-Leibler divergence, Aspect-oriented dependency tree

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: Aspect-based sentiment analysis (ABSA) is a task that aims to determine the sentiment polarity of aspects by identifying opinion words. Recently, numerous studies based on Graph Convolutional Networks (GCN) are used for ABSA and they mainly utilize a dependency tree to extract syntactic information. However, not all relations in a dependency tree are necessary and different context words have distinct influence on aspects. Thus, effectively utilizing syntactic information from dependency tree remains a challenging problem. In this paper, we present Distance-based and Aspect-oriented Graph Convolutional Network (DAGCN) to address the aforementioned issue, which consists of tow GCNs. Firstly, we propose a novel function called Distance-based Syntactic Value (DSV) to measure the importance of different context words syntactically and eliminate noise in the dependency tree, and thus construct Distance-based Weighted Matrix (DWM) with it. Secondly, we introduce Aspect-Fusion Attention (AF) to focus on the context words crucial for aspects in global scope and combine DWM with AF to integrate local and global syntactic information simultaneously.Finally, the first GCN (AoGCN) is designed based on the combined result to capture syntactic features and the second GCN (SaGCN) is designed with self-attention mechanism to learn semantic information. Furthermore, the Kullback-Leibler (KL) divergence loss is utilized to ensure that the features learnt by AoGCN and SaGCN are distinct. Extensive experiments on three public datasets demonstrate that DAGCN outperforms state-of-the-art models and verify the effectiveness of the proposed architecture.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 2308