Syntactic denoising and multi-strategy auxiliary enhancement for aspect-based sentiment analysis

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Aspect-based sentiment analysis (ABSA) traditionally relies on syntactic features from dependency trees.
• Noisy or irregular sentence structures in online reviews hinder the performance of syntactic-based Graph Convolutional Network (GCN) models.
• Insufficient integration of syntactic and semantic information limits sentiment detection accuracy in existing ABSA methods.

Purpose of the Study:

• To propose a novel approach, Syntactic Denoising with Multi-strategy Auxiliary Enhancement (SDMAE), to overcome limitations in ABSA.
• To enhance the accuracy of sentiment polarity detection for specific aspect terms.
• To improve the integration of syntactic and semantic information for better ABSA performance.

Main Methods:

• Pruning dependency trees by focusing on sentiment-critical part-of-speech features to reduce noise.
• Implementing a Multi-channel Adaptive Aggregation Module (MAAM) with multi-head attention to fuse semantic and syntactic representations.
• Employing a multi-strategy task learning framework with sentiment lexicons and supervised contrastive learning.

Main Results:

• The proposed SDMAE approach significantly improves performance across four benchmark datasets.
• Demonstrated superior results compared to several state-of-the-art methods in aspect-based sentiment analysis.
• Effective noise reduction in dependency parsing and enhanced feature integration were key to performance gains.

Conclusions:

• SDMAE offers a robust solution for aspect-based sentiment analysis, particularly for noisy text data.
• The synergistic integration of denoised syntactic information and semantic features leads to substantial performance improvements.
• The developed methods provide a promising direction for future research in sentiment analysis and natural language understanding.