JKT: A joint graph convolutional network based Deep Knowledge Tracing

[jchanxtarov]

Basic Information

• Authors: Xiangyu Song, Jianxin Li, Yifu Tang, Taige Zhao, Yunliang Chen, Ziyu Guan
• Date: 2021/08
• Published By: Information Sciences

Link

https://www.sciencedirect.com/science/article/abs/pii/S0020025521009142?via%3Dihub

Overview

• Knowledge Tracing = to model students' learning states based on sequences of students' learning activities
• This study proposes the novel knowledge tracing task (framework: JKT) to add "exercise-to-exercise" and "concept-to-concept" relationships to Knowledge Graph. By adding these relations, high levels of information can be obtained and the accuracy of student performance prediction can be improved.
• Previous GCN analysis work focuses primarily on a single graph and is never used in the area of knowledge tracing. JKT is a novel framework that extends GCN to the KT domain and fuses two sub-graphs.

Others

• 損失関数に重み付けを採用，（生成モデルgと確率pの関係性を要確認）
• モデル自体に（意思決定への）説明可能性があるのではなく，埋め込み表現の位置関係分析や生徒ごとの習熟度観察などで解釈性の面での有用性を示唆

Reference (for understanding)