Automatic essay scoring is a machine learning task where we create a model that can automatically assess student essay answers. Automated essay scoring will be instrumental when the answer assessment process is on a large scale so that manual correction by humans can cause several problems. In 2019, the Ukara dataset was released for automatic essay scoring in the Indonesian language. The best model that has been published using the dataset produces an F1-score of 0.821 using pre-trained fastText sentence embedding and the stacking model between the neural network and XGBoost. In this study, we propose to use a simpler classifier model using a single hidden layer neural network but using a richer feature, namely BERT sentence embedding. Pre-trained model BERT sentence embedding extracts more information from sentences but has a smaller file size than fastText pre-trained model. The best model we propose manages to get a higher F1-score than the previous models on the Ukara dataset, which is 0.829.

No abstract

Students' attendance in class is often mandatory in education and becomes a benchmark for assessing students. Sometimes there are still fraudulent practices by students to achieve minimum attendance. From the administrative perspective, a paper-based presence system is potentially wasteful and extends the administrative stage because it requires manual recapitulation. This study aims to design a class attendance application based on facial pattern recognition, smile, and closest Wi-Fi. The method used in this research is a deep learning approach with CNN based architecture, FaceNet, to recognize faces. In addition to facial images, the system will also validate the attendance with location and time data. Location data is obtained from matching SSID from the database, and time data is taken when the user sends attendance data through API. This attendance system consists of three applications: web, mobile, and services installed on a mini-computer, which are integrated to sending attendance data to the academic system automatically. As confirmation, students are required to smile selfies to strengthen the validity of their presence. The testing model's accuracy results are 92.6%, while for live testing accuracy the model obtained 66.7%.

In the process of verifying Al-Quran memorization, a person is usually asked to recite a verse without looking at the text. This process is generally done together with a partner to verify the reading. This paper proposes a model using Siamese LSTM Network to help users check their Al-Quran memorization alone. Siamese LSTM network will verify the recitation by matching the input with existing data for a read verse. This study evaluates two Siamese LSTM architectures, the Manhattan LSTM and the Siamese-Classifier. The Manhattan LSTM outputs a single numerical value that represents the similarity, while the Siamese-Classifier uses a binary classification approach. In this study, we compare Mel-Frequency Cepstral Coefficient (MFCC), Mel-Frequency Spectral Coefficient (MFSC), and delta features against model performance. We use the public dataset from Every Ayah website and provide the usage information for future comparison. Our best model, using MFCC with delta and Manhattan LSTM, produces an F1-score of 77.35%

The application of the Qur'an for the memorizers in adding and maintaining their memorization continues to grow in number. No less than 200 digital Qur'an applications are available on mobile application providers. In addition, publications on the topic of the digital Qur'an in the last ten years have also increased. Through these applications and publications, it is an opportunity to find patterns and knowledge about current topics and features. Through this knowledge, it is hoped that it can be a recommendation for a better form of digital Al-Qur'an application system, especially providing features that affect increasing the ease and quality of memorizing the Qur'an. This paper aims to explore the application of the Qur'an specifically for memorizing and papers on the topic to provide these recommendations. The method used to get the paper using PRISMA. While the applications being reviewed are taken from the AppStore. As a result, 31 papers were reviewed and 12 main applications regarding the Qur'an for memorization were obtained. Through the answers to each research question, it can be used by subsequent researchers as well as by system developers in developing Al-Qur'an products for better memorization of tense.