Sentiment classification via improved feature selection using Boolean operator-based particle swarm optimization.

Area of Science:

• Natural Language Processing
• Machine Learning
• Computational Intelligence

Background:

• Sentiment analysis is crucial for opinion mining but faces challenges in high-dimensional text data due to feature redundancy.
• Effective feature selection is key to improving sentiment classification accuracy.

Purpose of the Study:

• To introduce a novel Boolean Operator-based Particle Swarm Optimization (BOPSO) algorithm for enhanced feature selection in sentiment classification.
• To improve the efficiency and accuracy of sentiment analysis models by addressing high-dimensional data challenges.

Main Methods:

• Developed BOPSO by integrating Boolean logic operators (Adder, Subtractor, XOR) into Particle Swarm Optimization (PSO) for binary feature selection.
• Evaluated BOPSO on nine benchmark sentiment datasets using five filter-based objective functions (Chi-Square, Correlation, Gain Ratio, Information Gain, Symmetrical Uncertainty).
• Assessed classification performance using Naïve Bayes, Support Vector Machine (SVM), and Artificial Neural Network (ANN) classifiers.

Main Results:

• BOPSO demonstrated an average accuracy improvement of 1.8% to 4.5% over state-of-the-art optimization techniques (DE, GWO, ABC, CS).
• Achieved up to 100% accuracy with ANN on the laptop dataset, showing superior precision, recall, and F1-score.
• Effectively reduced feature dimensionality while significantly boosting sentiment classification performance.

Conclusions:

• The proposed BOPSO algorithm is a highly effective method for feature selection in sentiment analysis.
• BOPSO offers significant improvements in classification accuracy and efficiency compared to existing optimization techniques.
• This approach holds promise for advancing sentiment analysis in various application domains.