DAGCN: Dual-Channel and Aspect-Aware Graph Convolutional Network for Aspect-Based Sentiment Analysis in Computational Social Systems
Wanneng Shu, Cao Zhai, Ke Yu · IEEE Transactions on Computational Social Systems · 2024
It is crucial to mine latent sentiments and opinions from comments on social media to comprehensively understand people's preferences and experiences. Aspect-based sentiment analysis aims to identify the sentiment polarities for specific aspects of a sentence. Mainstream methods generally struggle to process sentences with complex syntactic structures and multiple aspects of different sentiment polarities. To overcome this challenge, this article proposes a dual-channel and aspect-aware graph convolutional network (DAGCN) model, which fully utilizes the syntactic and semantic information to accurately capture the sentiment features for specific aspects. Specifically, a syntactic graph convolutional network module is designed to effectively learn syntactic information by constructing an aspect-oriented dependency tree and employing a gated aggregator. To reduce the semantic interference from multiple aspects, a semantic graph convolutional network module is developed to capture both local and global semantic correlations corresponding to specific aspects. Furthermore, a BiAffine module is integrated to facilitate the interaction between the syntactic and semantic information. Experiments on four benchmark datasets illustrate that our proposed DAGCN significantly outperforms state-of-the-art baselines.