Sentiment Analysis of COVID-19 Public Activity Restriction (PPKM) Impact using BERT Method

Covid-19 has grown rapidly in all parts of the world and is considered an international disaster because of its wide-reaching impact. The impact of Covid-19 has spread to Indonesia, especially in the slowdown in economic growth. This was influenced by the implementation of Community Activity Restrictions (PPKM) which limited community economic activities. This study analyzes the mapping of public sentiment towards PPKM policies in Indonesia during the pandemic based on Twitter data. Knowing the mapping of public sentiment regarding PPKM is expected to help stakeholders in the policy evaluation process for each region. The method used is BERT with IndoBERT specific model. The results showed the evaluation value of the IndoBERT f-1 score reached 84%, precision 86%, and recall 84%. Meanwhile, f-1 scores 70%, 72% precision, and 70% recall for evaluating the use of SVM. Multinominal Naïve Bayes evaluation shows an f-1 score of 83%, precision of 78%, and recall of 80%. In conclusion, the BERT method with the IndoBERT model is proven to be higher than classical methods such as SVM and Multinominal Naïve Bayes. © 2022 Seventh Sense Research Group®

Age Classification of Moviegoers Based on Facial Image Using Deep Learning

The number of moviegoers in Indonesia continues to rise year after year until 2019. However, due to the COVID-19 pandemic, most Indonesian cinemas were closed in early 2020. Moviegoers are increasingly turning to digital platforms to watch films. Based on the films shown, they can be divided into three categories: films for children, films for adolescents, and films for adults. A system that can automatically classify the faces of the audience based on their age category is required. Using Deep Learning, this study aims to classify the audience's age based on facial photos. The first stage involves collecting data from three datasets: All-Age-Face, FaceAge, and FGNET, which are then combined and relabeled based on age group. Preprocessing and hyperparameter testing were also performed. Finding the best learning rate and bottleneck layer is the goal of hyperparameter testing. The training process employs learning rete and the two best bottleneck layers with six models, namely MobileNet, MobileNetV2, VGG16, VGG19, Xception, and ResNet101V2. Global Average Pooling was added at the end of the layer in each model. The MobileNet model on two bottleneck layers yielded the best testing accuracy value of 85.44 percent in this study © 2022, Abba Suganda Girsang & Dewa Bagus Gde Khrisna Jayanta Nugraha;published by UIKTEN. This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 License