The Use of Machine Learning in Online Course Sentiment Analysis: A Case Study Post COVID-19 Pandemic

Resumo

The number of tweets on Twitter containing information related to online lectures in the post-pandemic Covid-19 period or after the government lifted the pandemic period has drawn many pros and cons. This research discusses the best level of accuracy of the three machine learning methods, namely Naïve Bayes (NB), Decision Tree (DT) and Support Vector Machine (SVM) in analyzing Twitter results related to online lectures. The data used is crawled data from Twitter using the keyword “Online Lecture” and 5,978 tweets were obtained after verification. Three category labels were used in this research, namely Positive, Negative and Neutral. Pre-processing was carried out using the stages of Data Cleansing, Tokenizing, Stopword, Normalization and Stemming. While the feature extraction stage uses the TF-IDF stage. The average accuracy results for the NB method are 78.02%, precision is 78.18%, recall is 78.02% and f1-score is 77.45%. DT obtained an average accuracy of 86%, precision is 89.14%, recall is 81.14% and f1-score is 82.31%. The average accuracy obtained by SVM is 89%, precision 90.97%, recall 87.31%, and the F1 score is 85.99%. Among the three algorithms, it is known that SVM achieves very good performance in online lecture sentiment analysis based on Twitter data.