Hierarchical human-like strategy for aspect-level sentiment classification with sentiment linguistic knowledge and...

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Aspect-level sentiment analysis is vital for fine-grained understanding but challenging in real-world applications.
• Deep learning methods have advanced sentiment analysis, yet human cognitive processes remain underexplored.
• Human reading involves hierarchical comprehension and logical thinking, offering potential improvements for sentiment classification.

Purpose of the Study:

• To propose a novel Hierarchical Human-like strategy for Aspect-level Sentiment classification (HHAS).
• To integrate human reading cognition, including comprehension and logical thinking, into sentiment classification models.
• To enhance the effectiveness of aspect-level sentiment analysis by mimicking human reading stages.

Main Methods:

• Developed HHAS, a model incorporating three modules: sentiment-aware mutual attention, aspect-specific knowledge distillation, and reinforcement learning-based re-reading.
• These modules align with human cognitive stages: pre-reading, active reading, and post-reading.
• Conducted extensive experiments on three widely used datasets to evaluate HHAS performance.

Main Results:

• HHAS demonstrated impressive performance across all tested datasets.
• The proposed model achieved state-of-the-art results, outperforming existing methods.
• Experimental validation confirmed the effectiveness of the hierarchical, human-like approach.

Conclusions:

• The HHAS model successfully integrates human reading cognition into aspect-level sentiment analysis.
• The proposed strategy offers a significant advancement in fine-grained sentiment classification.
• HHAS provides a new, effective approach for tackling real-world sentiment analysis challenges.