[Paper Review] EDUKG: a Heterogeneous Sustainable K-12 Educational Knowledge Graph

Bowen Zhao, Jiuding Sun+10 more|arXiv (Cornell University)|Oct 21, 2022

Advanced Graph Neural Networks4 citations

TL;DR

EDUKG is a large-scale, heterogeneous, and sustainable knowledge graph for K-12 education in China, built using a fine-grained interdisciplinary ontology with 635 classes and 1,759 properties. It enables interactive knowledge extraction from textbooks and supports dynamic, scalable maintenance via a generalized entity linking system, resulting in a 252M-entity, 3.86B-triplet knowledge graph publicly available for research and educational applications.

ABSTRACT

Web and artificial intelligence technologies, especially semantic web and knowledge graph (KG), have recently raised significant attention in educational scenarios. Nevertheless, subject-specific KGs for K-12 education still lack sufficiency and sustainability from knowledge and data perspectives. To tackle these issues, we propose EDUKG, a heterogeneous sustainable K-12 Educational Knowledge Graph. We first design an interdisciplinary and fine-grained ontology for uniformly modeling knowledge and resource in K-12 education, where we define 635 classes, 445 object properties, and 1314 datatype properties in total. Guided by this ontology, we propose a flexible methodology for interactively extracting factual knowledge from textbooks. Furthermore, we establish a general mechanism based on our proposed generalized entity linking system for EDUKG's sustainable maintenance, which can dynamically index numerous heterogeneous resources and data with knowledge topics in EDUKG. We further evaluate EDUKG to illustrate its sufficiency, richness, and variability. We publish EDUKG with more than 252 million entities and 3.86 billion triplets. Our code and data repository is now available at https://github.com/THU-KEG/EDUKG.

Motivation & Objective

• To address the lack of interdisciplinary, fine-grained, and sustainable modeling of K-12 educational knowledge and resources.
• To overcome the limitations of existing educational KGs that focus only on coarse-grained subject knowledge or higher education content.
• To enable dynamic, incremental maintenance of educational knowledge by integrating heterogeneous online data sources.
• To support diverse downstream applications such as intelligent tutoring, question answering, and learning management systems.
• To promote educational equity by providing open access to comprehensive, high-quality learning resources for under-resourced learners.

Proposed method

• Design of a fine-grained, interdisciplinary ontology with 635 classes, 445 object properties, and 1314 datatype properties to uniformly model knowledge, resources, and heterogeneous data in K-12 education.
• Development of a semi-automated, interactive methodology for extracting factual knowledge from K-12 textbooks using the defined ontology.
• Creation of a generalized entity linking system to dynamically index and link new heterogeneous online educational resources to entities in EDUKG.
• Implementation of a scalable data pipeline to extract and integrate knowledge from textbooks and online sources into a unified RDF-based knowledge graph.
• Deployment of a public platform (open.edukg.cn) with APIs for knowledge search, entity linking, and question answering to support research and application development.
• Open-sourcing of the EDUKG codebase, data, and toolkits under a CC BY-NC-SA 4.0 license to ensure long-term sustainability and community contribution.

Experimental results

Research questions

• RQ1 How can a heterogeneous and sustainable K-12 educational knowledge graph be designed to support interdisciplinary and fine-grained modeling of both knowledge and educational resources?
• RQ2 What methodology enables efficient and interactive extraction of factual knowledge from K-12 textbooks while maintaining consistency with a unified ontology?
• RQ3 How can a generalized entity linking system be built to support dynamic indexing and integration of diverse, evolving online educational data into an existing knowledge graph?
• RQ4 To what extent does EDUKG outperform existing educational KGs in terms of data sufficiency, richness, and variability across subjects?
• RQ5 How can a large-scale educational knowledge graph be sustainably maintained and extended over time with minimal manual intervention?

Key findings

• EDUKG contains over 252 million entities and 3.86 billion triplets, making it the largest known interdisciplinary K-12 educational knowledge graph to date.
• The ontology defines 635 classes and 1,759 properties (445 object + 1,314 datatype), enabling fine-grained and cross-subject modeling of educational content.
• The interactive textbook knowledge extraction method successfully captures factual knowledge with high precision, supporting scalable knowledge acquisition.
• The generalized entity linking system enables dynamic indexing of heterogeneous online resources, ensuring long-term sustainability and extensibility of the knowledge graph.
• The public platform open.edukg.cn provides open APIs for knowledge search, entity linking, and question answering, enhancing accessibility and usability for researchers and developers.
• EDUKG is released under a permissive open license (CC BY-NC-SA 4.0), with full code and data available on GitHub, enabling reproducibility and community-driven extension.