Twitter Sentiment Analysis of COVID-19 and Education in Saudi Arabia

Abstract

Coping with COVID-19 implications had led to a transition to e-learning and e-testing approaches, some of which are being conducted for the first time. To assess the adoption of testing alternatives in Saudi Arabia, this study adopted the case of the 2020 Achievement Test, which was conducted remotely for the first time since being introduced in 2000. The test is expected to be a fair and standard test for all students wishing to join universities within the country. However, did the new test settings provide the expected fairness? This study utilised sentiment analysis (SA) to answer this question from the Saudi public view. To apply SA methods, the study collected qualitative data from Twitter, creating a corpus of 30,440 data points on the Saudi context. The study evaluated the effectiveness of using automatic data annotation based on the semantics of the unigrams and bigrams of the words and the emojis used in the text. Furthermore, the study exploited the use of both deep learning LSTM and BiLSTM algorithms, and baseline machine learning SVM and RF to classify the sentiment from the large data set. The study concludes with a summary of the sentiment from this large textual data set and evaluation of the abovementioned classification algorithms in term of accuracy and f1-score in addition to models’ loss for the deep learning algorithms.