A global optimization approach to multi-polarity sentiment analysis.

Area of Science:

• Social Media Mining
• Natural Language Processing
• Machine Learning

Background:

• Sentiment analysis is crucial for social media mining, with performance reliant on feature selection and classification.
• Information Gain (IG) and Support Vector Machines (SVM) are key techniques, but joint optimization is underexplored.
• The efficacy of global optimization in sentiment analysis requires further investigation.

Purpose of the Study:

• To propose and evaluate a global optimization-based sentiment analysis (PSOGO-Senti) approach.
• To enhance sentiment analysis performance by optimizing feature selection (IG) and SVM parameters globally.
• To assess the robustness and effectiveness of PSOGO-Senti across different domains and polarity levels.

Main Methods:

• Utilized Particle Swarm Optimization (PSO) for global optimization of feature dimensions and SVM parameters.
• Employed Information Gain (IG) for feature selection and Support Vector Machines (SVM) as the classification engine.
• Evaluated the PSOGO-Senti model on two distinct Chinese sentiment analysis datasets.

Main Results:

• PSOGO-Senti significantly improved binary and multi-polarity Chinese sentiment analysis.
• The approach effectively identified and removed redundant/noisy features, selecting domain-specific subsets.
• Greater improvements were observed in more complex multi-polarity (five-polarity) tasks compared to simpler ones (two-polarity).
• PSOGO-Senti outperformed Genetic Algorithm (GA) and grid search, particularly for challenging multi-polarity problems.

Conclusions:

• PSOGO-Senti is an effective and robust approach for sentiment analysis across various domains.
• The method demonstrates superior performance in handling complex multi-polarity sentiment analysis tasks.
• Global optimization offers a promising direction for advancing sentiment analysis techniques.