Skripsi
AUTOMATED ESSAY SCORING UNTUK PENILAIAN JAWABAN ESAI BAHASA INDONESIA DENGAN INDOBERT EMBEDDING DAN FEEDFORWARD NEURAL NETWORK
Improving the quality of education can be supported by a more effective assessment system, one of which is Automated Essay Scoring (AES) for automatic essay evaluation. This study develops an Indonesian-language AES system using IndoBERT Embedding and a Feedforward Neural Network (FNN). The dataset used is the secondary dataset from the UKARA Challenge, developed by the NLP Research Team at Universitas Gadjah Mada, which has a limited number of data and an imbalanced class distribution (labels 1 and 0). Overall, the developed model, after being trained and evaluated on datasets A and B, achieved an F1-score of 0.767. On dataset A, the model trained using the SMOTE technique obtained an F1-score of 0.835 with a batch size of 16, epoch 7, and a learning rate of 1.26e-4. The best model on dataset B achieved an F1-score of 0.699 with a batch size of 64, epoch 4, and a learning rate of 5.7e-3. These results indicate that IndoBERT Embedding and FNN provide a reasonably good performance compared to the baseline provided by UKARA for the training set, although challenges remain regarding data imbalance and limited dataset size.