MSDC: Aspect-level sentiment analysis model based on multi-scale dual-channel feature fusion

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Aspect-level sentiment analysis requires fine-grained text processing.
• Existing models struggle with high-dimensional syntactic dependencies and feature extraction.
• Multiple opinion words introduce noise, complicating aspect-term sentiment understanding.

Purpose of the Study:

• To propose a novel aspect-level sentiment analysis model (MSDC) addressing limitations of existing single-channel approaches.
• To enhance feature extraction and sentiment understanding through multi-scale dual-channel fusion.
• To improve the accuracy and F1 score in fine-grained sentiment analysis.

Main Methods:

• Implemented a multi-scale dual-channel feature fusion approach.
• Utilized multi-head gated self-attention and graph neural network channels for enhanced feature representation.
• Introduced an adaptive feature fusion mechanism to dynamically adjust aspect-to-context weighting.
• Integrated data processing using a capsule network.

Main Results:

• The proposed MSDC model demonstrates superior effectiveness on public datasets.
• Significant improvements in accuracy and F1 value were observed compared to existing technologies.
• The model excels in fine-grained text sentiment analysis tasks.

Conclusions:

• The MSDC model effectively addresses limitations in aspect-level sentiment analysis.
• Multi-scale dual-channel fusion and adaptive weighting enhance sentiment understanding.
• The model offers a promising advancement for fine-grained sentiment analysis applications.