Learning to Reuse Distractors to Support Multiple Choice Question Generation in Education

Multiple choice questions (MCQs) are widely used in digital learning systems, as they allow for automating the assessment process. However, due to increased literacy of students and advent social media platforms, MCQ tests shared online, teachers continuously challenged create new questions, which is an expensive time-consuming task. A particularly sensitive aspect creation devise relevant distractors, i.e., wrong answers that not easily identifiable being wrong. This paper studies how a large existing set manually created distractors over variety domains, subjects, languages can be leveraged help creating MCQs, by smart reuse distractors. We built several data-driven models based on context-aware question distractor representations, compared them with static feature-based models. The proposed evaluated automated metrics realistic user test teachers. Both automatic human evaluations indicate consistently outperform approach. For our best-performing model, average 3 out 10 shown were rated high-quality performance benchmark, make it public, enable comparison between different approaches introduce more standardized evaluation benchmark contains 298 educational covering multiple subjects & 77k multilingual pool vocabulary future research.