Laboratory for Image and Media Understanding,Department of Advanced Information Technology, Kyushu University
Research
Learning Analytic（Past）
Real-time Analytics
Real-time analytics enables teachers to understand students' activities even during a lecture period. The fundamental methodologies such as collection, aggregation and analysis of a large scale of learning logs are developed.
Journals (Peer-reviewed)
- Atsushi Shimada, Shin’ichi Konomi, Hiroaki Ogata
Real-Time Learning Analytics System for Improvement of On-Site Lectures
Interactive Technology and Smart Education, Vol.15, No.4, pp.314-331, 2018.12
BibTeX
Summarization of Learning Materials
Summarized learning materials enhance the pre-understanding of contents via preview. Image processing and natural language processing are utilized to select important pages, and optimization function is solved to generate a digest version of original learning material.
Journals (Peer-reviewed)
- Atsushi Shimada, Fumiya Okubo, Chengjiu Yin, Hiroaki Ogata
Automatic Summarization of Lecture Slides for Enhanced Student Preview -Technical Report and User Study-
IEEE Transactions on Learning Technologies, Vol.11, No.2, pp.165-178, 2017.03
BibTeX
Performance Prediction
Academic performance prediction is helpful for both of teachers and students to understand the current learning status and how it will relate to the final performance. A large scale of learning logs is utilized to make a prediction model.
International Conferences (Peer-reviewed)
- Fumiya Okubo, Takayoshi Yamashita, Atsushi Shimada, Yuta Taniguchi, Shin'ichi Konomi
On the Prediction of Students’ Quiz Score by Recurrent Neural Network
Multimodal Learning Analytics Across Spaces Workshop (CrossMMLA), 2018.03
BibTeX
Recommendation
Recommendation of related learning materials is helpful for students to understand the contents more deeply and/or to extend the knowledge about the contents. Page-wise recommendation is realized through the analytics of learning materials.
International Conferences (Peer-reviewed)
- Keita Nakayama, Masanori Yamada, Atsushi Shimada, Tsubasa Minematsu, Rin-ichiro Taniguchi
Learning Support System for Providing Page-wise Recommendation in e-Textbooks
The Society for Information Technology and Teacher Education (SITE2019), 2019.03
BibTeX
Knowledge Map Analytics
Knowledge map enables teachers and students to understand the situation of knowledge acquisition. Analytics methodologies to integrate multiple knowledge maps, to explore similar knowledge maps, to visualize the analytics results are developed.
International Conferences (Peer-reviewed)
- Akira Onoue, Masanori Yamada, Atsushi Shimada, Rin-ichiro Taniguchi
The Integrated Knowledge Map for Surveying Students’ Learning
The Society for Information Technology and Teacher Education (SITE2019), 2019.03
BibTeX
Enhancing the prediction of the students’ performance by Neural Network Models
Students’ performance prediction is the collecting students’ activities data from the learning education system, then generates a prediction model for analysis data and prediction students’ learning level. Instructors can use the prediction result from a model for discovery and invent at-risk students early to prevent these students failed on course. In the practical, the generation of prediction model that can use in common is a difficult task because of the difference of lecture plan. For instance, the difference of course length, lecture materials, teaching style on each course. This will make a model complicated to analyze and predict. Furthermore, the early prediction is a challenge these problems since the accuracy of prediction was low in the earliest stage because of a little student activity data. Thus, this study will use the artificial neural network technique to generate a versatile model to explore the important features through prediction and improve early prediction for timely intervention.
International Conferences (Peer-reviewed)
- Sukrit Leelaluk, Tsubasa Minematsu, Yuta Taniguchi, Fumiya Okubo, Takayoshi Yamashita, Atsushi Shimada
LSTM with Attention Mechanism for Students’ Performance Prediction Based on Reading Behavior
The 5th Workshop on Predicting Performance Based on the Analysis of Reading Behavior (LAK23 Data Challenge), 2023.03
BibTeX - Sukrit Leelaluk, Tsubasa Minematsu, Yuta Taniguchi, Fumiya Okubo, Takayoshi Yamashita, Atsushi Shimada
Scaled-Dot Product Attention for Early Detection of At-Risk Students
IEEE International Conference on Teaching, Assessment and Learning for Engineering (TALE2022), pp.316-322, 2022.12
BibTeX - Sukrit Leelaluk, Tsubasa Minematsu, Yuta Taniguchi, Fumiya Okubo, Atsushi Shimada
Predicting student performance based on Lecture Materials data using Neural Network Models
The 4th Workshop on Predicting Performance Based on the Analysis of Reading Behavior (LAK22 Data Challenge), 2022.03
BibTeX
Enhancing the prediction of the students’ performance by Neural Network Models
Students’ performance prediction is the collecting students’ activities data from the learning education system, then generates a prediction model for analysis data and prediction students’ learning level. Instructors can use the prediction result from a model for discovery and invent at-risk students early to prevent these students failed on course. In the practical, the generation of prediction model that can use in common is a difficult task because of the difference of lecture plan. For instance, the difference of course length, lecture materials, teaching style on each course. This will make a model complicated to analyze and predict. Furthermore, the early prediction is a challenge these problems since the accuracy of prediction was low in the earliest stage because of a little student activity data. Thus, this study will use the artificial neural network technique to generate a versatile model to explore the important features through prediction and improve early prediction for timely intervention.
International Conferences (Peer-reviewed)
- Sukrit Leelaluk, Tsubasa Minematsu, Yuta Taniguchi, Fumiya Okubo, Takayoshi Yamashita, Atsushi Shimada
LSTM with Attention Mechanism for Students’ Performance Prediction Based on Reading Behavior
The 5th Workshop on Predicting Performance Based on the Analysis of Reading Behavior (LAK23 Data Challenge), 2023.03
BibTeX - Sukrit Leelaluk, Tsubasa Minematsu, Yuta Taniguchi, Fumiya Okubo, Takayoshi Yamashita, Atsushi Shimada
Scaled-Dot Product Attention for Early Detection of At-Risk Students
IEEE International Conference on Teaching, Assessment and Learning for Engineering (TALE2022), pp.316-322, 2022.12
BibTeX - Sukrit Leelaluk, Tsubasa Minematsu, Yuta Taniguchi, Fumiya Okubo, Atsushi Shimada
Predicting student performance based on Lecture Materials data using Neural Network Models
The 4th Workshop on Predicting Performance Based on the Analysis of Reading Behavior (LAK22 Data Challenge), 2022.03
BibTeX
Educational Data Analysis using Generative AI
The purpose of this study is to explore the potential of using generative AI to analyze educational data, comparing the performance of two large language models (LLMs), GPT-4 and text-davinci-003, with respect to different types of analysis. We also integrate the LangChain framework with the LLMs in order to obtain useful analysis insights for novice data scientists. In addition, we study the impact of using the OpenLA library to preprocess educational data for analysis.
The results show that the GPT-4 provides the best analysis when using data preprocessed by OpenLA and that reading time and student activity in digital textbooks are important in predicting student performance.
The results show that the GPT-4 provides the best analysis when using data preprocessed by OpenLA and that reading time and student activity in digital textbooks are important in predicting student performance.
International Conferences (Peer-reviewed)
- Abdul Berr, Sukrit Leelaluk, Cheng Tang, Li Chen, Fumiya Okubo, Atsushi Shimada
Educational Data Analysis using Generative AI
The 6th Workshop on Predicting Performance Based on the Analysis of Reading Behavior (LAK24 Data Challenge), 2024.03
BibTeX