A Method Based on Attention Mechanism using Bidirectional Long-Short Term Memory(BLSTM) for Question Answering
Seyed Vahid Moravvej, Mohammad Javad Maleki Kahaki, Moein Salimi Sartakhti, Abdolreza Mirzaei
Abstract
Question answering (QA) enables the system to answer questions automatically. In recent years, much research has been done in this area. In most methods, question and answer words are given equal importance, which leads to poor model performance. This paper proposed Attention-Based Bidirectional Long-Short Term Memory(BLSTM) to select the answer to the question. In our model, first, word embedding is trained in several different ways. Then, we consider two BLSTM networks for question and answer. The outputs of these two networks and the difference between them are connected and entered into a feed-forward neural network. Finally, this network assigns a score to a question-answer pair. We evaluate our proposed model on the English and Persian datasets about Covid-19. The experiments demonstrate that our model achieves better results than other compared methods.