A Frequency-Channel Enhanced Dynamic GCN for Aspect-Based Sentiment Analysis

Abstract

Aspect-based Sentiment Analysis (ABSA) is a fundamental task in natural language processing. Most existing methods rely on graph-based approaches with dependency trees, which often suffer from parsing noise and rigid static graph structures. To address these issues, we propose a Frequency-Channel Enhanced Dynamic Graph Convolutional Network (FCD-GCN). Unlike static models, our approach introduces a dynamic graph mechanism that iteratively updates edge weights across layers, enabling adaptive capture of aspect-context associations. Our method consists of three steps: layer-wise dynamic graph construction, frequency-channel feature enhancement, and sentiment prediction. Additionally, a dual-attention strategy combines a frequency-domain module using Fourier transforms to model long-range dependencies and a channel-aware self-attention module to reduce feature redundancy. Extensive experiments on benchmark datasets demonstrate that FCD-GCN consistently outperforms strong baselines, validating the effectiveness of dynamic graph refinement with frequency and channel-level modeling. The code is publicly available at [GitHub].

[PDF]