An Ensemble Framework for Imbalanced Arabic Text-Based Emotion Analysis

Text analysis involves extracting knowledge from textual data for various applications. Emotion analysis can be conducted through multiple methodologies and serves a diverse array of purposes. In contemporary society, the sharing of experiences on social media platforms has become increasingly prevalent. For instance, Twitter serves as a valuable data source for organizations seeking to assess public opinions, sentiments, and emotional responses. Both organizations and individuals are keen to leverage social media for understanding public sentiment, extracting emotions, and gauging perspectives on specific issues; however, the field of emotion detection has received re class=WordSection2> latively limited focus. Previous studies have primarily explored emotional classifications within the text, particularly in Arabic content. The imbalance in datasets containing Arabic texts adversely impacts the classification process's effectiveness. Consequently, this research introduces an ensemble learning framework aimed at addressing this challenge, employing the Synthetic Minority Oversampling Technique (SMOTE) to achieve data balance, alongside Support Vector Machine (SVM), Naive Bayes (NB), and K-Nearest Neighbors (KNN) algorithms for emotion analysis. The SemEval-2018 dataset was utilized to evaluate the performance of the proposed methodology. Experimental findings validate the efficacy of the proposed model, which enhances the existing standards in classifying Arabic tweets, achieving an accuracy of 87.51% based on F-measures. The results indicate that the proposed analytical approach significantly advances text-based emotion detection and analysis, proving effective for Arabic text emotion analysis.