Sentiment Analysis for YouTube Educational Videos Using Machine and Deep Learning Approaches

Abstract

YouTube is one of the most popular video-sharing platforms. Recently, the use of YouTube as an educational tool has increased due to the covid-19 pandemic. The quality of the educational videos is crucial in the learning process. Users' comments on educational videos can help to determine the quality of the videos. Such comments can be utilized using a natural language processing technique called Sentiment Analysis. This research proposes a model to perform sentiment analysis on YouTube Arabic educational videos using classical machine learning classifiers and deep learning models. There are six experiments were conducted on two types of datasets: imbalance and balanced datasets. In addition, three balancing techniques were utilized which are: oversampling, under-sampling, and SMOTE. The results show that the best accuracy obtained from all the experiments was 96% by the SVC, RF, and DL model using the oversampling and SOMTE techniques. The worst accuracy achieved by the ML classifiers was 74% by the KNN classifier using the under-sampling technique. Meanwhile, the worst accuracy achieved by the DL model was 89% for the manually balanced dataset using the bigram, trigram, and five-gram. Additionally, this research introduced a novel dataset based on Arabic YouTube educational videos. This model can be beneficial to many researches in data mining research area or related research areas.