.png){kind=icon}
.png){kind=icon}
Return
Knowledge graph-based cognitive learning with multi-fact reasoning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neunet.2026.108747.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
An intelligent education system is a technology-driven comprehensive education system that is designed to support personalized learning by analyzing interactions between students, learning tasks, and knowledge concepts. Cognitive diagnosis (CD) in an intelligent educational system aims to diagnose the cognitive state of a student from historical response data, and further predict the future performance of the student on the exercise. However, existing CD models still suffer from the following issues: (1) most models only consider the score interactions of student-exercise and exercise-concept, which are difficult to capture the complex interactions among student-exercise-concept; and (2) most models neglect the higher-order interactions among student-exercise-concept. To address these issues, we propose a Cognitive Diagnosis framework based on Signed Knowledge Graph with multi-fact reasoning (CD-SKG). Specifically, we model three core elements in each student response log–student, exercise, and relevant knowledge concept–as individual facts. Since each response has two possible outcomes, we further define these facts as signed facts to encode response valence. Each signed fact inherently contains three distinct entity types, enabling systematic exploration of higher-order relationships within it across three levels, which aligns with our tripartite interaction logic. To distinguish the response impact on student performance, we employ a dual-view hypergraph convolutional network and learn the features of different responses on a signed cognitive hypergraph. In addition, we aim to reveal the different higher-order relations in each response record, which are analyzed at two diagnosis levels by integrating the higher-order information among the three cognitive factors. Finally, we conduct extensive experiments on four real datasets and achieve optimal performance in comparison with eight state-of-the-art CD models. Our code is publicly available at: https://github.com/jianruichen/CD-SKG .
Keywords:
Cognitive Diagnosis
Knowledge Graph
Multi-fact Reasoning
Student-Exercise-Concept Interaction
Hypergraph Convolutional Network