Generation of multiple choice questions using word prediction

Abstract

In education, multiple-choice questions are useful for both teachers and learners because they are suited to a large number and wide range and are in a format that can be used in many fields. These questions are basically created by the teacher, but creating many and wide questions makes burdens on the teacher heavy. So, it is considered effective to use questions generated by the system. The goal of automatic multiple-choice question generation is to generate distractors that cannot be clearly identified as incorrect at first glance. While previous research has proposed a method for generating distractors using the Semantic Web, we propose a method using word prediction based on a deep learning model. Since word prediction consider surrounding words, we think that it is possible to generate the best choice for each sentence. In this study, we asked students multiple-choice questions generated by the proposed method and the conventional method, and conducted questionnaire surveys and analyzed of their answers. As a result, it was confirmed that the proposed method can generate more difficult questions than the conventional method.