Entropy-aware filtering and hierarchical attention: A hybrid neural framework for fine-grained user satisfaction...

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Data Science

Background:

• Massive growth in user-generated reviews presents challenges for extracting user sentiment and satisfaction.
• Traditional methods for analyzing review usefulness and user satisfaction have limitations.

Purpose of the Study:

• To develop a novel multi-dimensional satisfaction analysis framework.
• To integrate sentiment analysis with fuzzy decision-making for enhanced user satisfaction evaluation.
• To improve the accuracy and robustness of user emotion modeling from text data.

Main Methods:

• A sentiment analysis module with entropy-based information usefulness prediction (IUP), PERT-enhanced BERT encoders for hierarchical feature extraction, and a BERT-wwm-ext model with hierarchical attention for sentiment prediction.
• A fuzzy decision-making component utilizing attribute weighting and an enhanced aggregation ranking strategy.
• Experiments conducted on 18,559 user reviews from mobility travel service systems (MTSS).

Main Results:

• The proposed framework demonstrated superior performance compared to 12 baseline methods.
• Achieved performance improvements of 1.4%-5.6% in accuracy, F1-score, and AUC.
• Effectively addressed limitations of vote-based usefulness metrics and mitigated noise from redundant content.

Conclusions:

• The developed framework offers a robust approach for fine-grained user emotion modeling.
• Provides effective multi-attribute satisfaction evaluation in complex, real-world scenarios.
• Advances the field of sentiment analysis and user satisfaction assessment in the context of big data.