RCD: Relation Map Driven Cognitive Diagnosis for Intelligent Education Systems

Weibo Gao, Qi Liu, Zhenya Huang, Yu Yin, Haoyang Bi, Mu-Chun Wang, Jianhui Ma, Shijin Wang

Abstract

Cognitive diagnosis (CD) is a fundamental issue in intelligent educational settings, which aims to discover the mastery levels of students on different knowledge concepts. In general, most previous works consider it as an inter-layer interaction modeling problem, e.g., student-exercise interactions in IRT or student-concept interactions in DINA, while the inner-layer structural relations, such as educational interdependencies among concepts, are still underexplored. Furthermore, there is a lack of comprehensive modeling for the student-exercise-concept hierarchical relations in CD systems. To this end, in this paper, we present a novel Relation map driven Cognitive Diagnosis (RCD) framework, uniformly modeling the interactive and structural relations via a multi-layer student-exercise-concept relation map. Specifically, we first represent students, exercises and concepts as individual nodes in a hierarchical layout, and construct three well-defined local relation maps to incorporate inter- and inner-layer relations, including a student-exercise interaction map, a concept-exercise correlation map and a concept dependency map. Then, we leverage a multi-level attention network to integrate node-level relation aggregation inside each local map and balance map-level aggregation across different maps. Finally, we design an extendable diagnosis function to predict students' performance and jointly train the networks. Extensive experimental results on real-world datasets clearly show the effectiveness and extendibility of our RCD in both diagnosis accuracy improvement and relation-aware representation learning.
BibTeX
@inproceedings{gao2021rcd,
  author = {Weibo Gao and Qi Liu and Zhenya Huang and Yu Yin and Haoyang Bi and Mu-Chun Wang and Jianhui Ma and Shijin Wang and Yu Su},
  title = {{RCD: Relation Map Driven Cognitive Diagnosis for Intelligent Education Systems}},
  booktitle = {International ACM SIGIR Conference on Research and Development in Information Retrieval},
  year = {2021},
  doi = {10.1145/3404835.3462932},
}

Loading related content...

Figure