Enhanced Arabic Sentiment Analysis Framework Based On Multi-Layer Feature Selection And Ensemble Learning

Sentiment analysis is the process of classifying, analyzing, and extracting meaningful information from unstructured user-generated data on social media. One major problem of sentiment analysis is the high dimensionality of the feature space due to a large number of terms. To address this problem, this research aims to design and implement an enhanced learning architecture for Arabic sentiment analysis. First, this work proposes a multi-layer feature selection method based on an ensemble of feature selection methods and particle swarm optimization. Second, this work proposes an enhanced learning model based on a multi-layer feature selection method and the ensemble of learning models for sentiment analysis. Results obtained from the proposed method showed high performance with an F-measure of 92.84, outperforming the baseline methods. These findings reveal that the proposed ensemble machine learning method, along with the proposed multi-layer feature selection method, is considered one of the most suitable techniques for Arabic sentiment analysis.