Sentiment Analysis using Multi-objective Optimization-based Feature Selection Approach

Abstract

Polarity categorization affects sentiment analysis. Categorization is a fundamental knowledge discovery challenge. Categorization accuracy depends on data quality. Hence, data must be preprocessed to obtain meaningful information. In real-world applications, data is enormous, and many properties are duplicates or useless. The classification depends on feature selection. It finds the best data representation qualities. Feature selection reduces model training time. When unnecessary attributes are eliminated, models learn better. Combinatorial optimization makes feature selection harder. Feature selection balances decreasing features and enhancing classification performance. We propose two multi-objective optimization techniques for the feature selection. The particle swarm optimization (PSO) and Krill Herd Algorithm (KHA) are applied for optimal feature selection. The proposed model consists of key steps such as review pre-processing, multi-objective optimization-based feature selection, and supervised classification. The performance of both PSO-based and KHA-based models is evaluated using the two sentiment analysis datasets. The results show the efficiency of both models in terms of precision, recall, accuracy, and F1-score parameters.