SENTIMENT ANALYSIS OF THE NATIONAL COVID-19 VACCINATION PROGRAM ON TWITTER USING THE BIDIRECTIONAL ENCODER REPRESENTATION FROM TRANSFORMER

In Indonesia, the implementation of the national COVID-19 (Coronavirus disease of 2019) vaccination programmes has received criticism from various strata of society, especially through social media platforms such as Twitter. Therefore, Twitter can be used as a data source to analyze Indonesian public sentiment regarding the vaccination programme. Various classical machine learning methods exist for sentiment analysis, but these methods require complex feature engineering and do not focus on the importance of word order in a sentence. In this study, a deep learning model, bidirectional encoder representation from transformer (BERT), is used to overcome these problems by conducting experiments to determine the best dataset after pre-processing, the best hyper-parameter, and the best pre-trained model for BERT. The data used in this study were Indonesian Twitter data with a total of 3000 tweets. Our results demonstrate that BERT is suitable for performing sentiment analysis. In our experiments, BERT obtained better results than classical machine learning methods, with a precision of 86.2%, recall of 86%, f1-score of 86%, and accuracy of 86%. The results of the sentiment analysis performed in this study can be considered by the government in formulating policies related to the implementation of vaccination programmes. ICIC International ©2023.

An Efficient Masked Face Classifier Using EfficientNet

The widening spread of an infectious disease, namely COVID-19 (corona virus disease 2019) has become a serious health problem for countries around the world. As an effort to control the confirmed case of COVID-19, WHO advice people to wear a face mask to protect themselves from the spread of the coronavirus. However, the people's awareness of wearing face masks in certain location is still low, so an automatic masked face classifier model is needed. This research performes the masked face classification into two class (masked and not-masked) by using the EfficientNet pretrained model. The experiment show that the EfficientNet is able to achieve, not only the best accuracy, but also the most efficient model, compared to the other state-of-the-art model of CNN, such as ResNet-50 and Inception. This condition is achieved when using both limited dataset and larger dataset. EfficientNet is able to achieve the 99% accuracy with about 5x lower number of network parameters and 3x faster testing time than ResNet-50. Compare to Inception, EfficientNet is better in terms of accuracy and also efficiency.