Comparative Study of Knowledge Graph Models in Education Domain

Zineb Elkaimbillah

, Maryem Rhanoui

, Mounia Mikram

and Bouchra El Asri

IMS Team, ADMIR Laboratory, Rabat IT Center, ENSIAS, Mohammed V University in Rabat, Morocco

Meridian Team, LYRICA Laboratory, School of Information Sciences, Rabat, Morocco

LRIT Laboratory, Rabat IT Center, Faculty of Sciences, Mohammed V University in Rabat, Morocco

Keywords:

Knowledge Graph, Educational Knowledge Graph, Knowledge Graph Embedding, Knowledge Graph

Application

Abstract:

Knowledge graph (KG) technologies are improving Artiﬁcial Intelligence. It can effectively expand the

breadth of search results. Therefore, KGs continue to solve several problems in different domains, including

the education ﬁeld. The application of educational KGs to learning systems has recently been expanded due to

increased demand in the education sector and the importance of KGs application to learning systems. In this

article, we present the knowledge Graph approach, the methodology of KG development, and analyze each

step. Also, we discuss the popular KG Embedding models. We provide a comparative study of KG models in

the education ﬁeld.

1 INTRODUCTION

The coronavirus pandemic 2019 (COVID-19) had

an impact on several ﬁelds, especially the ﬁeld of

education. Therefore, efforts are being made to

force educational systems into this vital sector. The

complementarity of bottom-up (machine learning-

driven) and top-down (semantic and knowledge

graph) techniques is essential for representation

insights from such data. Recently, the techniques of

machine learning for the representation of knowledge

are rapidly improving. As a result, the ﬁeld

of education has expanded by integrating the data

models of knowledge graphs. This context is

provided in a linking framework between semantic

metadata, for data sharing and integration. For

example, in semantic-aware Question Answering

(QA) services, semantic information may be used to

improve search results. (Berant et al., 2013)(Fader

et al., 2014)(Yao and Van Durme, 2014), Information

Retrieval(Liu et al., 2018)(Liu and Fang, 2015),

and Recommender System(Bellini et al., 2017)(Wang

et al., 2019).

The objective of this article is to present the

general context of the Knowledge Graph and the

process of its development. On the other hand, we

hope to provide a comparative study of knowledge

graphs applied in the education domain. We identify

the speciﬁc use and application of the KG, different

data sources used, and the techniques and models for

integrating the KG.

We organize the article as follows. Section 2

presents deﬁnitions of knowledge graphs, followed

by a discussion of various notable approaches to

knowledge graph development by presenting the

knowledge extraction aspect among their concepts

and related approaches, and we provide popular KG

embedding models. Finally, a comparative study of

KG in education is presented in Section 3.

2 BACKGROUND

In this section, we will approach some deﬁnitions

of Knowledge Graph (KG), then we will present the

methodology of development of a KG, and then we

will deﬁne what is Knowledge Graph Embedding by

citing its most known models.

2.1 Knowledge Graph Deﬁnition

Before presenting the different deﬁnitions of

knowledge graphs, we start with a reminder of two

important concepts ”knowledge” and ”graph”. The

ﬁrst one means the understanding of something such

as facts, skills, or objects. According to other points

of view, knowledge is acquired from many different

data sources and in different ways. The second

concept is mathematically deﬁned as a structure of

Elkaimbillah, Z., Rhanoui, M., Mikram, M. and El Asri, B.

Comparative Study of Knowledge Graph Models in Education Domain.

DOI: 10.5220/0010733800003101

In Proceedings of the 2nd International Conference on Big Data, Modelling and Machine Learning (BML 2021), pages 339-344

ISBN: 978-989-758-559-3

339

objects in which pairs of objects are called vertices,

nodes, or points are linked together by edges, links,

or lines.

In 2012, Google proposed a Knowledge Graph, to

use a Graph in their search engine. Many authors have

proposed several deﬁnitions of this concept. Notable

among these have already been, formally,(Tran and

Takasu, 2019) deﬁnes the knowledge graph as a

relation r between the main entity h and the tail entity

t. This deﬁnition is not complete enough as it lacks

the semantic level.

On the other hand, (Xiong, 2018) declared the

semantic level, deﬁning KG as a data resource that

contains entities, each entity has its speciﬁc semantics

and meaning.

(Vidal et al., 2019) presented the knowledge graph

as a representation capable of building formal models

of levels of abstraction and facts of different types.

Another effort is made to deﬁne a KG was

proposed by (Bonatti et al., 2019). The authors

consider it like an organized schema and several

labeled relations with a formal meaning are available.

2.2 Knowledge Graph Development

Several data sources are used in the construction of

knowledge graphs, such as small and large databases,

text documents, web pages, and crowdsourced

statements, and in either a manual, automatic, or

semi-automatic way (Bonatti et al., 2019).

The methodology of creating a Knowledge Graph

consists of two major components:

The top-down section: covers the process of

modeling a speciﬁc ﬁeld and contains the steps below

: (Fensel et al., 2020):

• Mapping step: describes a set of predeﬁned

mapping rules on an incoming data map for a

speciﬁc domain.

• Annotation development: This phase involves

converting speciﬁed domain standards into a form

interface to make manual and semi-automated

knowledge acquisition procedures easier.

The bottom-up section: which explains the process

of annotation in a new domain and contains the steps

listed below (Fensel et al., 2020):

• knowledge acquisition: First, we start this

process by deﬁning the domain (e.g., education).

The investigation of the domain entails knowledge

extraction from data sources (structured, semi-

structured, and unstructured).

• Knowledge extraction: describe types (Entity,

Relation, Attribute), approaches, and tools aspects

(Zhao et al., 2018). In terms of knowledge

extraction approaches, it heavily involves NLP,

text mining, and Machine learning. Examples

of techniques used for entities recognition and

relations extraction are Conditional Random Field

(CRF)(Su et al., 2009), Machine Learning models

(e.g., SVM), (BiLSTM) (Huang et al., 2015),

Hidden Markov Models (HMM)(Morwal et al.,

2012).

• Knowledge fusion: is a process that builds an

ontology and assesses its quality in an iterative

way (Zhao et al., 2018).

• Knowledge storage: This step consists of using

two main types of storage, the ﬁrst one, based on

RDF (Resource Description Framework), and the

second one, based on graphical databases. But in

general, the storage is done in NoSQL databases

(Zhao et al., 2018).

2.3 Knowledge Graph Embedding

Knowledge graph embeddings (KGEs) are the

process of constructing propositional feature

representations of entities and relations in vectors

in a knowledge graph (Wang et al., 2017) to apply

numeric techniques resulting in scalable besides

effectiveness.

The knowledge graph embeddings are computed

to fulﬁll speciﬁc properties; i.e., they follow given

KGE models to calculate the spatial distance between

two entities for the type of relationship in the low

dimensional plunge vector space.

Several popular KGE models have been widely

studied :

• Translating Embeddings for Modeling Multi

relational Date (TransE): This is the ﬁrst

translation model to get the tail vector as near

to the sum of the head and relation vectors as

feasible. (Bordes et al., 2013).

• Knowledge Graph Embedding by Translating

on Hyperplanes(TransH): The goal of transH

is to reduce the model’s complexity and the

difﬁculty of training by dealing with all possible

relationships (Wang et al., 2014).

• Learning Entity and Relation Embeddings for

Knowledge Graph Completion(TransR): TransR

can cover the 1-to-1 relationship, by separating

the relational space from the entity space (Lin

et al., 2015).

There are also more examples of KGE models

include DistMult (Yang et al., 2014), Complex

Embeddings (ComplEx) (Trouillon et al., 2016),

Holographic Embeddings (HolE) (Nickel et al.,

BML 2021 - INTERNATIONAL CONFERENCE ON BIG DATA, MODELLING AND MACHINE LEARNING (BML’21)

340

2016), Convolutional 2D KG Embeddings (ConvE)

(Dettmers et al., 2018), Convolution-based model

(ConvKB) (Nguyen et al., 2017). All KGE models

have advantages and disadvantages, so it is reasonable

to expect that a discover a more efﬁcient way to

represent knowledge.

3 COMPARATIVE STUDY

KGs have widely been used to beneﬁt a variety of

applications related to learning systems and a wealth

of pedagogical data.

K12EduKG is a knowledge graphing system

for K-12 educational topics. The K12EduKG

system’s architecture, which is made up of

two components: the concept extraction and the

relationship identiﬁcation are both used to extract

educational concepts for speciﬁc subjects. It

takes Named Entity Recognition (NER) methods

on educational data using the CRF model and

probabilistic association rule mining (Chen et al.,

2018a).

Using the object data of internal control policy

documents, (Wang, 2020) proposed a knowledge

graph construction method for internal control in

higher education institutions.

(Chen et al., 2018b) provided a KG can be used

for school topics and online courses to educate and

learn. To extract instructional concepts, this study

employs recurrent neural network (RNN) models on

pedagogical data. To identify educational relations

that interconnecting instructional concepts based on

student performance, They used the probabilistic

association rule mining method.

(Aliyu et al., 2020) present an automated

approach based on a knowledge graph to address the

problem of tertiary institution departments manually

arranging course allocations, and handling the task of

Question Answering to education administrators. To

provide intelligent knowledge services, the system

stores data in a knowledge graph in RDF/XML

format.

The improvement of the quality of teaching and

basic concepts of some computer science disciplines

is the objective of (Qin et al., 2020). Building

and visualizing educational knowledge as research

content using a huge quantity of course-related

information from the database as an example.

It begins with obtaining database knowledge,

cleaning and preparing the acquired data, and

then utilizing several automatic or semi-automatic

technological techniques to extract information,

identify knowledge units from text datasets, and

then extract entities, attributes, and relations.

Finally, the collected knowledge is saved in neo4j to

achieve knowledge graph visualization, producing an

effective educational knowledge graph of database

discipline to aid smart education.

MathGraph (Zhao et al., 2019) is a signiﬁcant

effort in educational KG development for

automatically solving high school mathematical

exercises. It is constructed using the crowdsourcing

technique to represent various mathematical objects.

(Yao et al., 2020) concentrate on this issue

and propose a model for embedding learning in

educational knowledge graphs. They present a

structural and literal embedding representation based

on TransE and Bert. They use three GRUs to

combine both methods, also using Knowledge Fores

and Wikipedia as a data source for the construction of

the educational KG.

We summarize the most important points for

this comparative study in Table 1 that represent an

overview of the KG approach in education, citing the

models end techniques embedding used, data source,

evaluation measure(s), and the limitations that can be

an improvement in the future research.

4 DISCUSSION

Analysis of KG construction methodologies derived

from scholarly publications in the ﬁeld of education

reveals a connected set of limitations and ﬂaws related

to the KG data type as several approaches have been

collected as textual data and the disregard of using

visual data. On the other hand, the majority of these

research either did not deﬁne the method(s) utilized in

the KG building, and strategies for extracting entities

and relations or did not provide any information on

the algorithm(s) used to create the KG.

To assess and enhance this work, we aim to

respond to all the problems detected such as by

including a comprehensive method for knowledge

extraction, construction, and analysis of educational

graphs using state-of-the-art models of natural

language processing (NLP), namely named entity

extraction, relation extraction, and coreference

resolution and methodological extensions of KG-

based multimodal (textual and visual data) extraction

and analysis algorithms.

Comparative Study of Knowledge Graph Models in Education Domain

341

Table 1: An outline of KG models applied in the education ﬁeld

Ref KG Usage Models Data Source Evaluation

Measure(s)

Limitation

(Wang,

2020)

Information

Retrieval and

recommendation

pTransE

CNKI database,

internal control

policy documents

mean

Silhouette

• Limited data sources,

• KG embedding model

not signiﬁcant

(Aliyu

et al.,

2020)

Managing

and allocating

courses, Question

Answering

Adhoc

Educational

information

system,

Case study

• Lack of proper

evaluation,

• limited KG resources

(Qin

et al.,

2020)

Improvement of the

quality of teaching

and basic concepts

of some computer

science disciplines,

Question

Answering

BILSTM-CRF,

word2vec

Course-related

information of

the database

Measure

Recall

Precision

• Insufﬁcient evaluation

(Yao

et al.,

2020)

Educational

Technology and

Linked prediction

TransE and

BERT

Knowledge Forest,

Wikipedia

Mean Rank

Hits@10

• Application in a limited

context,

• Limited resources used

for KG construction

(Zhao

et al.,

2019)

Solving

mathematical

exercises in high

school

Complex,

Conic and

Solid

domain experts and

Crowdsourcing

F1 measure

• Application in a limited

context,

• KG construction

required a limited

number of resources

(Chen

et al.,

2018a)

system for building

knowledge graphs

for K-12

probabilistic

association rule

mining, CRF

Chinese

mathematics

curriculum

AUC

• Inadequate assessment

metric

• A restricted application

on the KG was

proposed.

(Chen

et al.,

2018b)

Teaching and

learning on school

subjects and online

courses institutions

RNN, mining

probabilistic

association

rules

Pedagogical data,

evaluation of

learning

AUC

Precision

Recall

F1 measure

• Limited demonstration

of the KG for

mathematics

construction ,

• Limited on textual data

BML 2021 - INTERNATIONAL CONFERENCE ON BIG DATA, MODELLING AND MACHINE LEARNING (BML’21)

342

5 CONCLUSIONS

Knowledge Graph (KG) has a big impact on the

education ﬁeld, practice, improvement of the quality

of teaching, and solving high school exercises.

Overall, we conclude that KGs are capable of

providing semantically organized data.

In this paper, We discussed how knowledge

graphs can be used in a variety of domains,

including Question Answering, Recommendation,

and Information Retrieval. Also, we presented a

background for the KG approach, which includes

KG deﬁnitions, two methods of knowledge graph

construction: top-down and bottom-up, and the

presentation of KG Embeddings models. A

comparison of different models of knowledge graphs

utilized in the ﬁeld of education was offered.

We intend to expand this research in the

future by incorporating educational applications and

methodological extensions of KG-based algorithms

for multimodal extraction and analysis.

REFERENCES

Aliyu, I., Kana, A., and Aliyu, S. (2020). Development of

knowledge graph for university courses management.

International Journal of Education and Management

Engineering, 10(2):1.

Bellini, V., Anelli, V. W., Di Noia, T., and Di Sciascio,

E. (2017). Auto-encoding user ratings via knowledge

graphs in recommendation scenarios. In Proceedings

of the 2nd Workshop on Deep Learning for

Recommender Systems, pages 60–66.

Berant, J., Chou, A., Frostig, R., and Liang, P. (2013).

Semantic parsing on freebase from question-answer

pairs. In Proceedings of the 2013 conference on

empirical methods in natural language processing,

pages 1533–1544.

Bonatti, P. A., Decker, S., Polleres, A., and Presutti,

V. (2019). Knowledge graphs: New directions

for knowledge representation on the semantic web

(dagstuhl seminar 18371). In Dagstuhl Reports,

volume 8. Schloss Dagstuhl-Leibniz-Zentrum fuer

Informatik.

Bordes, A., Usunier, N., Garcia-Duran, A., Weston, J., and

Yakhnenko, O. (2013). Translating embeddings for

modeling multi-relational data. In Neural Information

Processing Systems (NIPS), pages 1–9.

Chen, P., Lu, Y., Zheng, V. W., Chen, X., and Li, X. (2018a).

An automatic knowledge graph construction system

for k-12 education. In Proceedings of the Fifth Annual

ACM Conference on Learning at Scale, pages 1–4.

Chen, P., Lu, Y., Zheng, V. W., Chen, X., and Yang, B.

(2018b). Knowedu: a system to construct knowledge

graph for education. Ieee Access, 6:31553–31563.

Dettmers, T., Minervini, P., Stenetorp, P., and Riedel,

S. (2018). Convolutional 2d knowledge graph

embeddings. In Proceedings of the AAAI Conference

on Artiﬁcial Intelligence, volume 32.

Fader, A., Zettlemoyer, L., and Etzioni, O. (2014).

Open question answering over curated and extracted

knowledge bases. In Proceedings of the 20th ACM

SIGKDD international conference on Knowledge

discovery and data mining, pages 1156–1165.

Fensel, D., S¸ims¸ek, U., Angele, K., Huaman, E., K

arle, E.,

Panasiuk, O., Toma, I., Umbrich, J., and Wahler, A.

(2020). Knowledge Graphs. Springer.

Huang, Z., Xu, W., and Yu, K. (2015). Bidirectional

lstm-crf models for sequence tagging. arXiv preprint

arXiv:1508.01991.

Lin, Y., Liu, Z., Sun, M., Liu, Y., and Zhu, X.

(2015). Learning entity and relation embeddings

for knowledge graph completion. In Proceedings

of the AAAI Conference on Artiﬁcial Intelligence,

volume 29.

Liu, X. and Fang, H. (2015). Latent entity space: a

novel retrieval approach for entity-bearing queries.

Information Retrieval Journal, 18(6):473–503.

Liu, Z., Xiong, C., Sun, M., and Liu, Z. (2018). Entity-duet

neural ranking: Understanding the role of knowledge

graph semantics in neural information retrieval. arXiv

preprint arXiv:1805.07591.

Morwal, S., Jahan, N., and Chopra, D. (2012). Named

entity recognition using hidden markov model

(hmm). International Journal on Natural Language

Computing (IJNLC), 1(4):15–23.

Nguyen, D. Q., Nguyen, T. D., Nguyen, D. Q., and

Phung, D. (2017). A novel embedding model for

knowledge base completion based on convolutional

neural network. arXiv preprint arXiv:1712.02121.

Nickel, M., Rosasco, L., and Poggio, T. (2016).

Holographic embeddings of knowledge graphs. In

Proceedings of the AAAI Conference on Artiﬁcial

Intelligence, volume 30.

Qin, Y., Cao, H., and Xue, L. (2020). Research and

application of knowledge graph in teaching: Take the

database course as an example. In Journal of Physics:

Conference Series, volume 1607, page 012127. IOP

Publishing.

Su, K.-Y., Su, J., Wiebe, J., and Li, H. (2009). Proceedings

of the joint conference of the 47th annual meeting

of the acl and the 4th international joint conference

on natural language processing of the afnlp. In

Proceedings of the Joint Conference of the 47th

Annual Meeting of the ACL and the 4th International

Joint Conference on Natural Language Processing of

the AFNLP.

Tran, H. N. and Takasu, A. (2019). Analyzing

knowledge graph embedding methods from a multi-

embedding interaction perspective. arXiv preprint

arXiv:1903.11406.

Trouillon, T., Welbl, J., Riedel, S., Gaussier,

E., and

Bouchard, G. (2016). Complex embeddings for

simple link prediction. In International Conference

on Machine Learning, pages 2071–2080. PMLR.

Comparative Study of Knowledge Graph Models in Education Domain

343

Vidal, M.-E., Endris, K. M., Jazashoori, S., Sakor, A., and

Rivas, A. (2019). Transforming heterogeneous data

into knowledge for personalized treatments—a use

case. Datenbank-Spektrum, 19(2):95–106.

Wang, H., Zhang, F., Zhao, M., Li, W., Xie, X., and Guo,

M. (2019). Multi-task feature learning for knowledge

graph enhanced recommendation. In The World Wide

Web Conference, pages 2000–2010.

Wang, J. (2020). Knowledge graph analysis of internal

control ﬁeld in colleges. Tehni

cki vjesnik, 27(1):67–

72.

Wang, Q., Mao, Z., Wang, B., and Guo, L. (2017).

Knowledge graph embedding: A survey of

approaches and applications. IEEE Transactions

on Knowledge and Data Engineering, 29(12):2724–

2743.

Wang, Z., Zhang, J., Feng, J., and Chen, Z. (2014).

Knowledge graph embedding by translating on

hyperplanes. In Proceedings of the AAAI Conference

on Artiﬁcial Intelligence, volume 28.

Xiong, C. (2018). Text representation, retrieval, and

understanding with knowledge graphs. PhD thesis,

Ph. D. Dissertation. Carnegie Mellon University.

Yang, B., Yih, W.-t., He, X., Gao, J., and Deng, L.

(2014). Embedding entities and relations for learning

and inference in knowledge bases. arXiv preprint

arXiv:1412.6575.

Yao, S., Wang, R., Sun, S., Bu, D., and Liu, J.

(2020). Joint embedding learning of educational

knowledge graphs. In Artiﬁcial Intelligence Supported

Educational Technologies, pages 209–224. Springer.

Yao, X. and Van Durme, B. (2014). Information extraction

over structured data: Question answering with

freebase. In Proceedings of the 52nd Annual Meeting

of the Association for Computational Linguistics

(Volume 1: Long Papers), pages 956–966.

Zhao, T., Chai, C., Luo, Y., Feng, J., Huang, Y., Yang,

S., Yuan, H., Li, H., Li, K., Zhu, F., et al. (2019).

Towards automatic mathematical exercise solving.

Data Science and Engineering, 4(3):179–192.

Zhao, Z., Han, S.-K., and So, I.-M. (2018). Architecture

of knowledge graph construction techniques.

International Journal of Pure and Applied

Mathematics, 118(19):1869–1883.

BML 2021 - INTERNATIONAL CONFERENCE ON BIG DATA, MODELLING AND MACHINE LEARNING (BML’21)

344