2020 Fiscal Year Annual Research Report
AI enhanced adaptive tutoring system by generating individualized questions and answers based on cognitive diagnostic assessment
| Project/Area Number |
20J15339
|
|---|
| Research Institution | National Institute of Information and Communications Technology |
|---|
Principal Investigator |
GAN Wenbin 国立研究開発法人情報通信研究機構, ユニバーサルコミュニケーション研究所統合ビッグデータ研究センター, 研究員
|
|---|
| Project Period (FY) |
2020-04-24 – 2022-03-31
|
| Keywords | Knowledge Assessment / Cognitive Diagnosis / Knowledge Tracing / Performance Modeling / Item Response Theory / Intelligent Tutoring / AI in Education / Education Data Mining |
|---|
| Outline of Annual Research Achievements |
In this year, I focused on the dynamic learner's knowledge assessment to model learner performance and discover learner's latent knowledge states. I have proposed two models to trace learner knowledge from their exercising logs accumulated in the online learning systems.
One is published in an international journal Applied Intelligence to handle the issue of insufficient learning factor modeling. It investigates the learner factors (learning and forgetting) and domain factor (item difficulty) by making use of rich information during learners’ learning interactions and proposes a novel model that traces the evolution of learners’ knowledge acquisition over time by explicitly modeling their learning and forgetting behaviors as well as the item difficulty. Extensive experiments confirmed the effectiveness of this model.
The other is published in an international conference BESC 2020 for the oral presentation. It achieves the goal of fine-grained assessment and interpretability. This novel model unifies the strength of the deep memory network to represent the separate knowledge states and the interpretability of the Item Response Theory (IRT) to explain the learner performance. Extensive experiments demonstrate the superiority and interpretability of the model for learner performance modeling.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
This research is generally going well beyond expected. It proposes two models for tracing learner knowledge states dynamically based on the learners’ exercising logs on the online learning systems. These two models model leaners’ performance and overcome issues of existing models from different perspectives.
One explores the factors that influence the knowledge acquisition and makes use of rich information during learners’ learning interactions to achieve more precise prediction of learner knowledge. The other explores to combine the knowledge tracing and cognitive diagnosis assessment to fulfill the requirements of both large-scale assessment and interpretability to explain the diagnosed results. Both models are evaluated on the real-world datasets and demonstrate the superiority over existing models. The research results are published in one international journal and an intentional conference, making good contributions to the research field.
The above work implements the idea in the research plan. This is a big improvement to the existing framework, and will also be a starting point to some future works. I also strengthen myself a lot and grow up as a researcher smoothly through this work.
|
| Strategy for Future Research Activity |
I will continue my research plan, and contribute more to the intelligent tutoring community. Firstly, I will further explore the current research of fine-grained assessment and interpretability. We have verified its effectiveness on learner performance modeling, however, the deep model analysis to investigate its ability and the effectiveness of its modules has not been fully conducted. I will further test the ability of knowledge proficiency inference in a multi-granularity manner and the interpretability of the learner performances in terms of the knowledge states and item characteristics.
Secondly, I will keep on the research in learner knowledge assessment, especially paying attention to some fundamental issues of the existing work. Especially, the performances of existing models greatly suffer from the sparseness of the input data and the information loss when modeling the learning process. How to alleviate the sparseness and the information loss while improving the model performance is another future work. We will develop new techniques to cope with this issue.
Thirdly, the work of learner knowledge assessment can be further improved by incorporating multimodal data collected during the learning process. This multimodal learning analysis will build more reliable model for assessing the learners.
|
Research Products
(2 results)
