EDUKG a Heterogeneous Sustainable K-12 Educational Knowledge Graph Bowen Zhao1 Jiuding Sun23 Bin Xu13 Xingyu Lu3 Yuchen Li3 Jifan

2025-05-03 0 0 4.3MB 17 页 10玖币

侵权投诉

EDUKG: a Heterogeneous Sustainable K-12

Educational Knowledge Graph

Bowen Zhao1?, Jiuding Sun2,3?, Bin Xu1,3( ), Xingyu Lu3, Yuchen Li3, Jifan

Yu3, Minghui Liu3, Tingjian Zhang3, Qiuyang Chen3, Hanming Li3, Lei Hou3,

and Juanzi Li3

1Global Innovation Exchange, Tsinghua University, Beijing, China

{zhaobw21,sjd22}@mails.tsinghua.edu.cn

2Khoury College of Computer Sciences, Northeastern University, Boston MA, USA

3Department of Computer Science and Technology,

Tsinghua University, Beijing, China

xubin@tsinghua.edu.cn

Abstract. Web and artiﬁcial intelligence technologies, especially se-

mantic 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 het-

erogeneous sustainable K-12 Educational Knowledge Graph. We ﬁrst

design an interdisciplinary and ﬁne-grained ontology for uniformly mod-

eling 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 inter-

actively 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 dynam-

ically index numerous heterogeneous resources and data with knowledge

topics in EDUKG. We further evaluate EDUKG to illustrate its suﬃ-

ciency, richness, and variability. We publish EDUKG with more than

252 million entities and 3.86 billion triplets. Our code and data reposi-

tory is now available at https://github.com/THU-KEG/EDUKG.

Keywords: Ontology ·Knowledge Graph ·K-12 Education

1 Introduction

The object of education is to prepare the young to educate themselves through-

out their lives, as said by Robert M. Hutchins. Education, especially for K-12

children, plays a signiﬁcant role in everyone’s life. Intelligent education, which

aims to leverage the Web and artiﬁcial intelligence (AI) technologies to improve

students’ learning eﬃciency [13,19], has always been an essential topic for re-

searchers. In addition, the construction of educational knowledge graphs (KGs)

?Equal Contributions.

arXiv:2210.12228v1 [cs.CL] 21 Oct 2022

2 B. Zhao et al.

is a fundamental research with variable downstream applications, including edu-

cational data mining [22], learning management systems [1], question answering

platforms [30], dialogue systems [18], etc.

Plentiful educational KGs have been proposed to help the development of

computer-aided educational technologies. KnowEdu [4] and K12Edukg [3] are

constructed by extracting concepts and prerequisite rules from subject-speciﬁc

textbooks. However, entities in these KGs are only course concepts without

other essential educational resources for students. CKGG [23] is proposed based

on Chinese high-school-level geography education, yet they only integrate data

for location entities. Meanwhile, several educational KGs are proposed based

on massive online open courses (MOOCs). For instance, MOOC-KG [6] and

HEKG [34] are built upon open course data, yet their ontology can only repre-

sent subject-speciﬁc knowledge at a shallow level. In particular, there are only 4

and 6 deﬁned classes in MOOC-KG and HEKG, respectively. Furthermore, al-

though KGs built upon open courses consist of heterogeneous data, they cannot

dynamically develop with growing resources. Additionally, most KGs based on

MOOCs are designed for higher education instead of K-12 education.

Despite that several KGs have been proposed for educational usage, they

suﬀer from the following limitations:

•Insuﬃcient Knowledge Modeling. Prior research pointed out that in-

terdisciplinary teaching is beneﬁcial for developing students’ critical thinking,

creativity, communication, and essential academia [12]. In the meantime, ﬁne

knowledge granularity is also beneﬁcial for students’ learning process [25]. Nev-

ertheless, existing educational KGs only represent subject-speciﬁc knowledge on

a coarse-grained level, lacking interdisciplinary entity relations.

•Sophisticated Data Curation. Education aims to teach students with

broad ability instead of just knowledge in textbooks [28]. Educational resources,

including examination questions and beyond, are proved to be beneﬁcial for

fostering students’ abilities through learning by doing [2,24]. Also, existing data

repositories for education, such as MOOCCUBEX [32] leverages a concept graph

to organize heterogeneous data altogether. However, existing educational KGs

still lack adequate resources due to data heterogeneity.

•Neglected Information Growth. Information for education is ever-growing

from both knowledge and data perspectives. For knowledge, the educational

reform in China is consistently changing the essential knowledge of education

through time. For data, there are increasing online materials for students to

learn. Nonetheless, prior educational KGs lack maintenance sustainability, i.e.,

the ability to capture and infuse new knowledge and resources incrementally.

To address these issues, we conclude that educational KGs should be built

with an interdisciplinary schema that can represent not only knowledge but also

resources. Meanwhile, towards maintaining sustainability, an educational KG

should be able to grow and adapt incrementally to the change of real-world

knowledge. Therefore, we propose EDUKG, a heterogeneous sustainable K-12

Educational Knowledge Graph for Chinese high-school-level education. We de-

sign an interdisciplinary ﬁne-grained ontology that uniformly models knowledge,

EDUKG: a Heterogeneous Sustainable K-12 Educational Knowledge Graph 3

Fig. 1: EDUKG’s data suﬃciency compared with other KGs. Blue, yellow, red,

and green points refer to knowledge and resources in interdisciplinary, geography,

politics, and history subjects, respectively.

resources, and heterogeneous data. In total, we deﬁne 635 classes and 1759 prop-

erties in EDUKG ontology without subject boundaries. Guided by this ontology,

we propose a semi-automated method for interactively acquiring knowledge from

textbooks. Fig. 1 compares data in EDUKG with Wikidata1and CKGG, indi-

cating that EDUKG consists of most suﬃcient information from both knowledge

and data perspectives. Additionally, for sustainably maintaining EDUKG with

growing data, we propose a general mechanism to index heterogeneous online

data incrementally based on our proposed entity linking technique.

Contributions. In general, our contributions are summarized as follows:

1. An interdisciplinary, ﬁne-grained ontology uniformly represents K-12 educa-

tional knowledge, resources, and heterogeneous data with 635 classes, 445

object properties, and 1314 datatype properties;

2. A large-scale, heterogeneous K-12 educational KG with more than 252 mil-

lion entities and 3.86 billion triplets based on the data from massive educa-

tional and external resources;

3. A ﬂexible and sustainable construction and maintenance mechanism empow-

ers EDUKG to evolve dynamically, where we design guiding schema of the

construction methodology as hot-swappable, and we simultaneously monitor

32 diﬀerent data sources for incrementally infusing heterogeneous data.

Outline. In the following sections, we ﬁrst illustrate the ontology for EDUKG

in Sec. 2, and we present EDUKG construction and maintenance mechanisms

in Sec. 3. Afterward, in Sec. 4, we introduce essential characteristics of EDUKG

to prove its suﬃcient qualities. In Sec. 5, we present the impact and availability

of EDUKG with its data, code, and applications. Finally, the related works are

introduced in Sec. 6, and we conclude our paper in Sec. 7

2 Schema of EDUKG

In this section, we introduce the ontology of EDUKG, which uniformly represents

knowledge, resource, and heterogeneous data.

1https://www.wikidata.org

4 B. Zhao et al.

Fig. 2: An overview of EDUKG top-level ontology.

2.1 Overview of EDUKG Ontology

As shown in Fig. 2, we divide EDUKG ontology into three main sections, which

are “Knowledge Topic”, “Educational Resource”, and “External Heterogeneous

Data”. Here we deﬁne the three top-level classes in EDUKG as follows:

–“Knowledge Topic”: essential themes in some speciﬁc subjects [10] and their

essential rhetorical roles.

–“Educational Resource”: intra-curricular teaching and testing resources in

K-12 education, for example, textbooks and examination exercises.

–“External Heterogeneous Data”: extra-curricular resources and data give

students vast approaches to learning more comprehensive knowledge.

Since EDUKG contains educational knowledge and resources, we investi-

gate and follow multiple published knowledge and resource modeling standards.

For knowledge, we reuse vocabularies from the widely-adopted RDF and RDFS

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

EDUKG:aHeterogeneousSustainableK-12EducationalKnowledgeGraphBowenZhao1?,JiudingSun2;3?,BinXu1;3(),XingyuLu3,YuchenLi3,JifanYu3,MinghuiLiu3,TingjianZhang3,QiuyangChen3,HanmingLi3,LeiHou3,andJuanziLi31GlobalInnovationExchange,TsinghuaUniversity,Beijing,Chinafzhaobw21,sjd22g@mails.tsinghua.edu.cn2Khour...

展开>> 收起<<

EDUKG a Heterogeneous Sustainable K-12 Educational Knowledge Graph Bowen Zhao1 Jiuding Sun23 Bin Xu13 Xingyu Lu3 Yuchen Li3 Jifan.pdf

共17页,预览4页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

EDUKG a Heterogeneous Sustainable K-12 Educational Knowledge Graph Bowen Zhao1 Jiuding Sun23 Bin Xu13 Xingyu Lu3 Yuchen Li3 Jifan

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: