Dual syntax aware graph attention networks with prompt for aspect-based sentiment analysis.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Aspect-based sentiment analysis (ABSA) is complex due to multiple aspects and polarities in sentences.
• Pre-trained language models (e.g., BERT) and graph neural networks (GNNs) show promise but have limitations.
• Existing methods struggle with irregular syntax and underutilize advanced pre-trained models.

Purpose of the Study:

• To propose a novel Dual Syntax-aware Graph attention networks with Prompt (DSGP) model for ABSA.
• To address limitations of existing methods in handling complex sentence structures and leveraging pre-trained models.
• To improve the accuracy and robustness of aspect-based sentiment analysis.

Main Methods:

• The DSGP model employs prompt templates to maximize pre-trained model potential.
• It uses masked vector outputs from templates as supplementary aspect features.
• It integrates dependency and constituent trees with graph attention networks for comprehensive syntactic information extraction.

Main Results:

• The proposed DSGP model demonstrates competitive performance across four public datasets.
• Fusion of prompt outputs and syntactic information from dual trees enhances feature representation.
• The model effectively captures both fine-grained word correlations and high-level sentence structure.

Conclusions:

• The DSGP model offers a significant advancement in aspect-based sentiment analysis.
• Leveraging dual syntax trees and prompt-based pre-trained models improves performance.
• The approach shows promise for handling complex linguistic phenomena in sentiment analysis.