KnowEdu: A System to Construct Knowledge Graph for Education

TLDR

The authors propose KnowEdu, a system that automatically constructs educational knowledge graphs to meet growing demands in the education domain. KnowEdu extracts instructional concepts from pedagogical data using neural sequence labeling and identifies educational relations from learning assessment data via probabilistic association rule mining, integrating heterogeneous sources to build the graph. Evaluation shows concept extraction achieves an F1 score above 0.70, while relation identification attains an AUC of 0.95 and MAP of 0.87.

Abstract

Motivated by the vast applications of knowledge graph and the increasing demand in education domain, we propose a system, called KnowEdu, to automatically construct knowledge graph for education. By leveraging on heterogeneous data (e.g., pedagogical data and learning assessment data) from the education domain, this system first extracts the concepts of subjects or courses and then identifies the educational relations between the concepts. More specifically, it adopts the neural sequence labeling algorithm on pedagogical data to extract instructional concepts and employs probabilistic association rule mining on learning assessment data to identify the relations with educational significance. We detail all the above mentioned efforts through an exemplary case of constructing a demonstrative knowledge graph for mathematics, where the instructional concepts and their prerequisite relations are derived from curriculum standards and concept-based performance data of students. Evaluation results show that the F1 score for concept extraction exceeds 0.70, and for relation identification, the area under the curve and mean average precision achieve 0.95 and 0.87, respectively.