Laboratory for Image and Media Understanding,Department of Advanced Information Technology, Kyushu University
Research
Data Analysis
DiaRoBERTa: A Multi-Party Dialogue Model for Multi-Skill Recognition in Classroom Collaborative Problem Solving Discussions
This study proposes DiaRoBERTa, a model for CPS dialogue classification. It uses MPC techniques, [SPK]/[THN] markers (encoding speaker roles/transitions) and focal loss (addressing class imbalance). Tested on 40 middle school chemistry discussion transcripts, it hits 85.51% accuracy, aiding educational CPS skill assessment.
International Conferences (Peer-reviewed)
- Shuqing LIU, Li CHEN, Sijie Xiong, Haiqiao LIU, Cheng TANG, Atsushi SHIMADA
DiaRoBERTa: A Multi-Party Dialogue Model for Multi-Skill Recognition in Classroom Collaborative Problem Solving Discussions
The 1st International Conference on Learning Evidence and Analytics (ICLEA 2025) , 2025.09
BibTeX
A Distilled Model for Collaborative Problem Solving Skill Classification on Resource-Limited Devices
This study develops a knowledge distillation framework for lightweight, accurate Collaborative Problem Solving (CPS) classification on resource-limited devices. It uses DiaRoBERTa (teacher) and DistilBERT (student) with two markers, cuts model size by 32%, achieves 81.16% accuracy, outperforms MobileBERT/ALBERT, aiding real-time educational evaluation.
International Conferences (Peer-reviewed)
- Shuqing LIU, Li CHEN, Haiqiao LIU, Cheng TANG, Fumiya OKUBO, Atsushi SHIMADA
A Distilled Model for Collaborative Problem Solving Skill Classification on Resource-Limited Devices
The 1st International Conference on Learning Evidence and Analytics (ICLEA 2025) , 2025.09
BibTeX