AEGA: enhanced feature selection based on ANOVA and extended genetic algorithm for online customer review analysis

Area of Science:

• Natural Language Processing
• Machine Learning
• Data Mining

Background:

• Sentiment analysis aims to understand user opinions from online platforms for performance improvement.
• High-dimensional feature sets in online reviews pose challenges for accurate classification.
• Existing feature selection techniques struggle to achieve high accuracy with minimal features.

Purpose of the Study:

• To develop an effective hybrid approach for sentiment analysis using an enhanced genetic algorithm (GA) and analysis of variance (ANOVA).
• To achieve high classification accuracy with a significantly reduced number of features.
• To overcome the local minima convergence problem in classification models.

Main Methods:

• A hybrid approach combining an enhanced genetic algorithm (GA) with analysis of variance (ANOVA) was developed.
• A unique two-phase crossover and selection strategy was employed to enhance exploration and convergence.
• ANOVA was utilized to drastically reduce the feature size, minimizing computational burden.

Main Results:

• The proposed hybrid approach achieved 78.60% accuracy and 79.38% F1 score on the Amazon Review dataset.
• On the Restaurant Customer Review dataset, the model obtained 77.70% accuracy and 78.24% F1 score.
• The novel approach outperformed existing algorithms, utilizing approximately 45% and 42% fewer features for the respective datasets.

Conclusions:

• The hybrid GA-ANOVA approach is effective for sentiment analysis, offering high accuracy with reduced feature sets.
• The method demonstrates superior performance compared to conventional classifiers and optimization algorithms.
• This approach effectively minimizes computational load while maintaining high classification performance in review analysis.