Research Interest

A unit is said to be a guideline that defines the relevant content and prerequisite knowledge to be learned [Contractor et al., 2015]. In Japan, uniform standards known as the Courses of Study are often used, and most schools and companies have adopted them. However, many digital teaching materials either do not include unit information or use proprietary classifications. To conduct systematic analysis across schools and individuals, common information must be used regardless of the teaching materials and services used. 

In the educational field, digital educational materials uploaded in all layouts are difficult to extract complete text [Abekawa & Aizawa, 2016], and it is difficult to achieve high accuracy with a contextual analysis approach. Therefore, we constructed an algorithm that can accurately assign unit information automatically even when the textual information extracted from various PDF materials is incomplete, by using n-grams as features, which are known to be able to handle incomplete sentences, and showed that it can automatically assign units with an accuracy of more than 91% [Yamauchi et al., 2023]. 

Current recommendation systems provide students with a selection of existing materials that the system chooses as appropriate. Still, the range of materials is limited and does not always meet the student's needs. There is a need to provide more individualized and optimized materials in real time, in line with students' errors, with feedback and explanations of the reasons for their errors. 

In our doctoral research, we will focus on generating individually optimized analogous questions. It is believed that having students reflect on their errors [Heemsoth & Heinze, 2016], especially by creating and solving analogous problems to the ones they answered incorrectly, promotes student understanding. Still, it is said to be less effective for students who lack experience in creating analogous problems [Sakitani, 1996]. This study generates analogous questions with individualized explanations using the students' response data. Furthermore, taking a cue from the correspondence between individuality and individually optimized messages found in previous research [Yamauchi et al., 2023], we aim to provide answers and explanations to content in a format that suits each student by including the student's performance and personality in the prompts.  

Ensuring accuracy and dealing with erroneous output must be considered in educational practice, especially in computational tasks where the generating AI produces erroneous output [Frieder et al., 2023]. Many models have been proposed in which AI predicts students' knowledge states based on the outcomes of their efforts [Vie et al., 2019]. However, AI has never been able to learn in the process of receiving students' questions from other students and improving learning materials in order to generate better teaching materials. 

In this study, we build a system to recommend learning materials that are effective for learning to other students, using the question generation system completed in the above research. First, we will refer to the logs and reviews of students who have actually worked on the generated problems to determine whether the problems should be recommended or not. We will also create a mechanism to request evaluations from higher-graders and teachers, allowing them to ask questions when questions arise in solving the problems. Evaluators can use the modification mechanism to improve questions in an interactive manner, depending on the situation. The next question to be recommended is then recommended to different students in similar learning situations. The evaluated data is also used as labeled data to train the AI. 

I am interested not only in producing research results but also in how research results can be applied to education in the field. For this reason, I am challenging myself to take a teaching course to experience education in the field first-hand, and I am thinking daily about what research results can be linked to schools by actually conducting classes using digital teaching materials from the standpoint of a teacher.

I have also continued to study mathematics in secondary education after entering university, as I believe that one must have a good knowledge of the topics being researched, not only as a researcher but also as a teacher. I have been teaching a wide range of students from primary school to high school for five years at a cram school called "Risusha", and I have also obtained a Grade pre-1 Suuken Test.