Determining Fuzzy Membership for Sentiment Classification: A Three-Layer Sentiment Propagation Model.

Area of Science:

• Natural Language Processing
• Machine Learning
• Sentiment Analysis

Background:

• Vast amounts of review data require effective sentiment analysis for consumers and manufacturers.
• Existing sentiment scoring methods lack precision, often relying on lexicons or feature locations.
• Sentiment scores offer more detailed insights than simple polarity (positive/negative).

Purpose of the Study:

• To propose a novel three-layer sentiment propagation model (TLSPM) for more accurate review sentiment scoring.
• To enhance sentiment classification effectiveness and reduce prediction errors in review analysis.
• To integrate document sentiment intensity into classification models.

Main Methods:

• Developed a three-layer sentiment propagation model (TLSPM) utilizing interrelations among documents, topics, and words.
• Employed nine pairwise relationship matrices to model sentiment propagation.
• Used fuzzy membership from TLSPM as weights for a fuzzy support vector machine (FSVM) model.

Main Results:

• The proposed TLSPM achieved more accurate sentiment scores for review documents.
• FSVM trained with TLSPM demonstrated enhanced sentiment classification effectiveness compared to SVM.
• FSVM trained with TLSPM reduced the mean square error (MSE) on sentiment rating prediction datasets.

Conclusions:

• TLSPM provides a robust framework for sentiment analysis by considering document, topic, and word interrelations.
• Weighting text based on fuzzy document membership derived from TLSPM improves classification performance.
• The approach offers significant improvements in both sentiment classification accuracy and prediction error reduction.