Zylich, Brian, Adam Viola, Brokk Toggerson, Lara Al-Hariri, and Andrew Lan. “Exploring Automated Question Answering Methods for Teaching Assistance.” In Artificial Intelligence in Education, edited by Ig Ibert Bittencourt, Mutlu Cukurova, Kasia Muldner, Rose Luckin, and Eva Millán, 610–22. Lecture Notes in Computer Science. Cham: Springer International Publishing, 2020. https://doi.org/10.1007/978-3-030-52237-7_49.
One important aspect of learning is through verbal interactions with teachers or teaching assistants (TAs), which requires significant effort and puts a heavy burden on teachers. Artificial intelligence has the potential to reduce their burden by automatically addressing the routine part of this interaction, which will free them up to focus on more important aspects of learning. We explore the use of automated question answering methods to power virtual TAs in online course discussion forums, which are heavily relied on during the COVID-19 pandemic as classes transition online. First, we focus on answering frequent and repetitive logistical questions and adopt a question answering framework that consists of two steps: retrieving relevant documents from a repository and extracting answers from retrieved documents. The document repository consists of course materials that contain information on course logistics, e.g., the syllabus, lecture slides, course emails, and prior discussion forum posts. This question answering framework can help virtual TAs decide whether a question is answerable and how to answer it. Second, we analyze the timing of student posts in discussion threads and develop a classifier to predict the timing of follow-up posts. This classifier can help virtual TAs decide whether to respond to a question and when to do so. We conduct experiments on data collected from an introductory physics course and discuss both the utility and limitations of our approach .