SHORT ANSWER GRADING IN BAHASA USING BERT

• Main Author: Putri Dinanti, Anissa
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/56338
• Institution: Institut Teknologi Bandung

— Short answer grading is the type of question that aim to gain understanding of students. Type of question asked in short answer grading is objective question and the answer has one phrase to one paragraph length. Short answer question has three types of question, there are definition question, explanation question, and giving example question. BERT is the state-of-the-art of various NLP tasks. The BERT technique that is the focus of the research is siamese BERT, fine tuning BERT, and auxiliary BERT. In siamese BERT, sentence embedding is done, which is a method of converting a sentence into a vector. The similarity of the vectors is seen by using the calculation of cosine similarity in the linear regression model. The fine tuning BERT method is used to modify BERT into sentence pair classification task. In auxiliary BERT, the method used also uses BERT for sentence pair classification task with additional auxiliary sentences in question sentences and data labels equipped with pseudo-sentences. Meanwhile, the pre-trained models used were DistilBERT, IndoBERT, and XLM-RoBERTa