Laboratory for Image and Media Understanding,Department of Advanced Information Technology, Kyushu University
Research
Learning Analytics
Ranking-based at-risk student prediction using federated learning and differential features
This study proposes a novel method for predicting academic performance that balances both privacy protection and high prediction accuracy. The approach integrates federated learning, which enables multiple educational institutions to collaboratively train models without sharing raw data, and a technique that represents student performance using relative differential features rather than absolute scores.
By doing so, it becomes possible to protect student privacy while mitigating the decline in prediction accuracy caused by differences in data distributions across institutions.
Furthermore, the method generates a risk ranking based on the predicted results, ordering students by their likelihood of academic underperformance. This function allows educators to efficiently identify and prioritize students who require early intervention and support.
By doing so, it becomes possible to protect student privacy while mitigating the decline in prediction accuracy caused by differences in data distributions across institutions.
Furthermore, the method generates a risk ranking based on the predicted results, ordering students by their likelihood of academic underperformance. This function allows educators to efficiently identify and prioritize students who require early intervention and support.
International Conferences (Peer-reviewed)
- Shunsuke Yoneda, Valdemar Švábenský, Gen Li, Daisuke Deguchi, Atsushi Shimada
Ranking-Based At-Risk Student Prediction Using Federated Learning and Differential Features
18th International Conference on Educational Data Mining, 2025.07
BibTeX
The Effect of Feedback in Chatbot-based Pre-class Learning Environment
In this work, with the rapid advancement of LLMs, chatbots support pre-class learning but may cause misunderstandings. Exercises with feedback can help. This study developed chatbot-based tools and compared two feedback types—KCR and ExF—examining their distinct effects on learning outcomes and effectiveness in promoting early awareness and deeper understanding.
International Conferences (Peer-reviewed)
- Vimeanseth Thorng, Fumiya Okubo, Atsushi Shimada
The Effect of Feedback in Chatbot-based Pre-class Learning Environment
The 1st International Conference on Learning Evidence and Analytics (ICLEA 2025) , 2025.09
BibTeX
Integrating Scaffolding Strategies with Environmental Monitoring Systems to Enhance Learning and Practical Skills in Agricultural Education
Scaffolding supports students by guiding them step-by-step toward mastering complex concepts. In agricultural education, students often struggle with abstract topics like humidity or soil composition due to limited real-world context. To address this, we introduce a field environment digest system that connects theoretical knowledge with field data. This study proposes a teaching approach combining scaffolding strategies and the digest system to enhance agricultural learning. A quasi-experimental design is applied in a high school setting. The experimental group uses the combined approach, while the control group follows traditional methods. We collect data through questionnaires measuring students’ learning interest, confidence, and skill development. Results show that the experimental group scores significantly higher on post-tests. Students also report greater motivation and confidence in understanding agricultural topics. The findings suggest that integrating scaffolding with real-world systems improves both academic performance and engagement.
International Conferences (Peer-reviewed)
- Haiqiao LIU, Tsubasa MINEMATSU, Chengjiu YIN, Shuqing LIU, Sijie XIONG & Atsushi SHIMADA
Integrating Scaffolding Strategies with Environmental Monitoring Systems to Enhance Learning and Practical Skills in Agricultural Education
The 1st International Conference on Learning Evidence and Analytics (ICLEA 2025) , 2025.09
BibTeX
Fine-tuned T5 Models on FairytaleQA Chinese Dataset
Question Answering (QA) is very important for comprehension learning and FairytaleQA is widely employed in this domain. However, rare versions in a limited number of alphabet languages restricts its application and current translators have five fatal errors. In our study, we manually translate FairytaleQA into Chinese and test its effectiveness via five fine-tuned T5 models.
International Conferences (Peer-reviewed)
- Sijie Xiong, Haoling Xiong, Tao Sun, Haiqiao Liu, Fumiya Okubo, Cheng Tang, Atsushi Shimada
Fine-tuned T5 Models on FairytaleQA Chinese Dataset
The 1st International Conference on Learning Evidence and Analytics (ICLEA 2025) , 2025.09
The Best Poster Award
BibTeX
Enhancing Attention-Based Knowledge Tracing with Digital Textbook Interaction
This study proposes an enhanced Knowledge Tracing model that integrates digital textbook viewing logs with exercise response data. By weighting reading time with content similarity, the model more accurately predicts learners’ knowledge states. Experiments on university course data show improved performance over traditional approaches, emphasizing the value of incorporating broader learning behaviors.
International Conferences (Peer-reviewed)
- Kotaro Kawabata, Fumiya Okubo, Yuta Taniguchi, Cheng Tang, Atsushi Shimada
Enhancing Attention-Based Knowledge Tracing with Digital Textbook Interaction
The 1st International Conference on Learning Evidence and Analytics (ICLEA 2025) , 2025.09
BibTeX