Accurately predicting student performance is a key challenge for adaptive learning systems. In realistic lecture-based environments, accurately modeling learners' understanding is difficult due to the sparsity of assessment opportunities and the lack of extensive prior response data. While deep learning-based approaches, such as Deep-IRT, achieve high predictive accuracy, they rely heavily on large-scale historical datasets, limiting their practical applicability. Furthermore, achieving a balance between predictive performance and model interpretability remains a long-standing challenge. To address this gap, we propose an interpretable learning analytics framework that integrates multimodal behavioral signals into an extended Item Response Theory (IRT) model. By utilizing facial expression and gaze data collected via Head-Mounted Displays (HMDs) in an XR-based lecture environment, we estimate a time-varying latent student state parameter (ϕ)utilizing Multi-Layer Perceptron (MLP) and Temporal Convolutional Network (TCN) models. This approach compensates for the dynamic learning states—such as engagement and cognitive load—that traditional static response logs fail to capture. We evaluated the proposed models with 20 students across three experimental scenarios simulating varying amounts of available response data, including the complete absence of pre-test data. The experimental results demonstrate that under sparse data conditions, the MLP model effectively learns robust behavioral patterns and outperforms the complex TCN model. Crucially, even without any historical pre-test data, the MLP model utilizing behavioral signals consistently outperformed the traditional IRT model and achieved a high predictive accuracy of 0.840, nearly matching the performance when full historical data is available. These findings indicate that multimodal biometric data can significantly compensate for missing prior knowledge assessments, establishing a highly efficient and practical foundation for realizing personalized learning without the time costs of extensive pre-testing.

Understanding whether students have understood specific knowledge points during lectures is a central challenge for adaptive learning systems. This study investigates whether short-term changes in students' effective ability can be used to detect knowledge-point-level misunderstandings in XR-based lectures. Each quiz item is mapped to a corresponding knowledge point, and each knowledge point is aligned with the lecture interval in which it is explained. Using multimodal behavioral data, including facial expressions and gaze, collected from 20 students in an XR-based lecture environment, we estimate a behavior-adjusted effective ability within an extended Item Response Theory (IRT) framework. The effective ability is defined as the sum of a static baseline ability and a behavior-derived state offset. By comparing effective ability before and after each knowledge-point interval, we infer whether the student has understood the corresponding concept. Results show that ability-change-based detection achieves 74.2% accuracy in identifying misunderstandings across 10 target knowledge points, with particularly strong performance for conceptual distinction tasks (78.5%). Error pattern analysis reveals three common types of misconceptions: concept confusion, tacit knowledge gaps, and definition recognition failures. A preliminary targeted intervention based on these findings yields significant learning gains (mean improvement 32.7%, p < 0.05). These findings suggest that short-term changes in behavior-adjusted effective ability can provide an interpretable basis for knowledge-point-level misunderstanding detection in data-sparse XR classroom conditions.

In this paper, we propose a system that defines and visualises learners' understanding of lecture content based on learning data obtained from the digital textbook system, Smart E-textbook Application (SEA), and a quiz function. By feedbacking the visualised results to the learner, the system allows the learner to know which units of the lecture need a review. This enables the learner to improve their understanding of lectures efficiently and effectively. To evaluate whether the proposed system can improve learning effectiveness and learners' understanding of lectures, we conducted an evaluation experiment and found that the proposed system does both. Additionally, a clear difference in the number of logs in the digital textbook and the time to answer the quiz was confirmed. From these, it is possible to decide whether the learner is actively engaged in learning or not.

This research proposal describes the development of a virtual space platform aimed at transforming learning and collaboration in the post-corona era. This research uses technology and the Unity platform to construct a space where learning data can be collected and where students attend lectures that are close to reality.

To cope with Japan's rapid population aging and declining birthrate, local governments in Japan are required to promote DX and develop DX human resources. In response, I developed a survey to measure the DX literacy level of local government officials. Since various questions were required, I used generative AI to create the questions, analyzed the results of question generation using IRT, and confirmed that generative AI was influential in creating skill survey questions in the DX literacy field.

Currently, local governments in Japan need DX (=Digital Transformation) promotion personnel. However, even today, Japan's civil service system still employs mainly legal and human personnel, resulting in a need for more necessary human resources. Therefore, the demand for DX human resource development for local government officials is rapidly increasing. In response to this, we have been engaged in human resource development using e-learning materials based on Moodle for three years since 2020, targeting approximately 9,500 employees of the Hiroshima City Office, where I work, and through formative evaluation of the materials, we have succeeded in creating materials at a level where all employees who took the course expressed a desire to promote DX. The company succeeded in developing educational materials so that all staff members who took the course were willing to promote DX. However, only about 5% of all staff members completed the course, reaching the limit of the improvement method in formative evaluation from those who completed the course. During the three-year effort, we also identified a wide gap in the IT skills of local officials and that we cannot provide optimal educational materials for 95% of the staff. In this study, we created multiple-choice questions corresponding to the DX literacy standard set by the Ministry of Economy, Trade and Industry (METI) in December 2022, which we regarded as a complete mastery map, and optimized them using IRT. The objective was to develop a model for measuring the DX literacy of local government officials and an e-learning course that provides individualized and optimal learning based on the measurement results. Through these efforts, we would like to establish a method of correlating skill surveys using IRT (=Item Response Theory) with e-learning course attendance data to overcome the limitations of the technique of creating e-learning courses and improving them through formative evaluation and to present a method to overcome the limitations of formative assessment in the ID (=Instructional Design) field. We hope to offer a strategy to overcome the limitations of formative evaluation in ID.

Recently, digital textbook systems and various exercise support systems have been introduced to educational institutions, and research has been conducted to collect and analyze logs in order to improve teaching and learning. In addition, research has been conducted to incorporate group work into exercises in order to prevent differences in the level of understanding in conceptual modeling lectures, and effective pairing methods have been proposed based on the learners' level of understanding. In the previous studies on pairing methods, logs of learning behaviors such as the results of exercises and the status of reading textbooks as well as dialogues were analyzed and surveyed. Therefore, in this study, we developed a chatbot system that supports learners in preventing the modification of their work and correcting it by incorporating a third party's opinion based on the learner's understanding and discussion status into the discussion.

Traffic accidents at intersections are the most common type of traffic accidents in Japan, and practicing driving at intersections is essential to prevent traffic accidents. However, practicing driving in a real car carries risks such as traffic accidents. Therefore, by using VR, driving practice can be done safely and realistically. In addition, the eye tracking technology in the HMD can be used to confirm whether the driver was able to see traffic signs and pedestrians while driving and whether the driver was following the correct procedures, leading to improved driving skills. In this study, we recreated an intersection where traffic accidents frequently occur in a VR space, practiced driving using eye tracking technology, and collected data on driving maneuvers and eye movement.

The government has mandated online education due to the recent COVID-19, whereas in-person lectures have been increasingly restricted. Videotelephony and online chat platforms like as Zoom and Microsoft Teams are frequently used to conduct online classes. However, expressions and reactions are more difficult to observe online than in person. With the advancement of virtual reality technology, we can construct virtual education environments where users can communicate and interact using virtual avatars. Of course, the risk of contracting COVID-19 will be eliminated by adopting this environment. In this study, I would like to offer an education system that enables students to attend lectures by manipulating avatars in a virtual reality classroom using an electronic textbook. In addition, I chose to employ a VR helmet with eye tracking to overcome the problem of receiving expressions and reactions so slowly. By collecting and analyzing eye tracking data from students during lectures, the system will identify the issues encountered by the majority of students and give graphs to assist teachers in enhancing the quality of instruction.

This paper describes a design of virtual reality classroom toward Kansei engineering analysis. This study explores how to design the atmosphere and environment that make it easy to concentrate on class and to increase the sense which take a lecture in online using virtual reality technology. Kansei Engineering is appealing to the world under the name of Affective Engineering [1].

This paper describes the effects of a pairing method based on digital textbook logs and learners’ artifacts in conceptual modeling exercises. We developed a digital textbook system called Smart E-textbook Application (SEA) and a conceptual modeling tool called KIfU 3.0 to collect conceptual modeling activity logs in exercises. This study proposes a method that makes pairs of learners for group work by considering the characteristics of the artifacts created by them and digital textbook logs. An initial evaluation was conducted to evaluate the learning effects of our proposed pairing method compared to the random pairing method. From its results, this study found the discussion patterns and digital textbook browsing status in the maximum value of improvement and deterioration points of learners’ artifacts in the conceptual modeling.

This paper proposes a system to define and visualize learners’ level of understanding of lecture contents based on learning data obtained from the digital textbook system named SEA (Smart E-Textbook Application) with a quiz function. By feedbacking the visualization results to learners, learners can grasp information that should be reviewed which unit in the lecture. Thereby, the learners can improve their level of understanding of the lecture efficiently and effectively. An evaluation experiment was conducted to investigate whether our proposed system can enhance learning effectiveness and learners’ level of understanding of lecture. As a result, it found our proposed system can enhance learners’ level of understanding of lecture and learning effectiveness.

When practicing car driving in a real world, there are risks like traffic accidents. With the development of virtual reality technology, we can create environments in which people can practice without being exposed to certain dangers like traffic accidents that potentially occur in real environments. So far, many researches regarding car driving in VR have been developed, but little attention has been paid to analyzing and visualizing eye gaze data collected in VR education systems. Analyzing and visualizing the collected eye gaze data is potential to grasp information of whether learners overlooked objects such as traffic signals and signs. In this study, we developed a driving simulator in VR and collected eye tracking data from five university students while they were practicing driving using eye tracking technology. Also, we have developed a system that allows learners to reflect on their own driving using the collected their eye gaze data. We compared the changes in gaze between those who received feedback to improve their driving skills and those who did not. The results showed that the feedback was effective because it reduced the number of oversights while driving.