End-to-End Question Answering System for Indonesian Documents Using TF-IDF and IndoBERT

In this paper, we propose a Question Answering System (QAS) for Indonesian text using retriever-reader architecture. The retriever is designed to retrieve relevant documents given a question, while the reader goal is to extract answers from those documents. We built the retriever based on TF-Idf model that returns most relevant documents based on cosine similarity score. On the other hand, we built the reader by fine-tuning IndoBERT for question answering task. We performed the evaluation on translated SQuAD dataset, with Exact Match and F1-score as the metrics. The evaluation results show that the proposed system demonstrates good performance, achieving the highest F1-score of 70% and the highest Exact Match score of 53% on the translated SQuAD dataset.