Roberta-based Encoder-decoder Model for Question Answering System

Abstract

Question answering is an essential component of the chatbot where it can facilitate users retrieve tremendous information from online websites. Traditional question answering system are built upon the conventional deep learning sequence models, where they are failed to capture sufficient semantic features due to the complexity of the question answering task. To tackle the above challenges, we propose a novel approach, equipped with a Roberta-based encoder-encoder framework, to extract the underlying semantic features more effectively. The experimental results on a well-studied dataset, i.e., SQUAD, show that our proposed method outperforms the baseline models in term of both EM and F1 evaluation metrics, which proves our model effectiveness.