Automatic Question Generation for Chatbot Development
Ryusei Doi, Thatsanee Charoenporn, Virach Sornlertlamvanich · 2022 7th International Conference on Business and Industrial Research (ICBIR) · 2022
It is a labor intensive task to prepare a list of questions for the intents in creating a chatbot system. It is not so easy to predict the variation of questions to be matched with the answers or what we want to provide the information to the chatbot users. In this research, we aim to solve the problem by applying a transformer framework to generate a question sentence from a list of keywords and explanatory text. In this paper, the question text is generated by using a trained Japanese T5 model by applying a list of keywords extracted from the questions and the explanatory text of the Amagasaki City FAQ database. To expand the coverage of the questions in matching the chatbot intent, we apply WordNet to expand the keywords in the questions. Finally, Semantic Textual Similarity (STS) Sentence-BERT is applied to measure the similarity of the user query and the question list in the chatbot.