Arabic Sentiment Analysis towards Feelings among Covid-19 Outbreak Using Single and Ensemble Classifiers

Abstract

The need to study and analyze public opinions about the Corona virus (COVID-19) pandemic or about those preventive measures that are imposed, led to the emergence of many studies. These conducted studies have concerned the analysis of public feelings and opinions, known as sentiment analysis (SA). Taking a benefit of social media platforms such as Twitter a dataset of Arab people feelings, especially fear and anxiety, towards Covid-19 was built through surveying the Arabic content in this platform. A machine learning (ML) model was applied to analyze and categorize the tweets related to fear and anxiety regarding Covid-19 outbreak. In this model, the word2vec was employed for word embedding to form the vector of features with two CBOW pre-trained models CC.AR.300 and Arabic.news. Moreover, the effect of the sampling technique that is called Synthetic Minority Over-sampling Technique and Edited Nearest Neighbors (SMOTENN) was investigated in this study. In addition, the performance of several single-based and ensemble classifiers were evaluated and discussed. The experimental results show that applying word embedding and SMOTENN with both single and ensemble classifiers achieve a good improvement in terms of F1 average score compared to the baseline, single and ensemble classifiers without SMOTENN.