以深度學習方法進行事實問題產生之研究

In an educational environment, quiz is still one of the most important teaching tools. Over the past few decades, the studies in the traditional classroom environment and large-scale open-line courses have found that it has better learning effect to provide learners with frequent and sufficient questions instead of spending only the same amount of time to study notes or textbooks. The goal of automatic question generation is to generate a natural language description from a given sentence or paragraph to test the learner's understanding on the knowledge content. The factual question is important in the assessment, which can assess the learner's comprehensive memory of declarative or factual knowledge. Besides, providing factual questions during the learning process allows learners to improve knowledge retention by stimulating the process of recall. The question generation system also can be used as a component in a chat bot, such as asking questions to start a conversation or requesting feedback, which provides companionship, educational function, or a clinical tool for assessing or improving mental health. Question generation techniques are also used to develop an annotated data set for computer reading comprehension and question answering. According to the importance of this research topic, this project will use Chinese textbooks in specific fields as the source documents. We will study how to apply the deep learning methods to develop the fundamental technologies for constructing a model for generating factual questions. The following research two subtopics will be studied: (1) how to extract important entities and relationships of facts from given documents, (2) how to generate factual questions by an end-to-end approach.