Year

2023

Degree Name

Master of Philosophy

Department

School of Computing and Information Technology

Abstract

The evolution of the Internet technology and artificial intelligence has changed the ways we gain knowledge, which has expanded to every aspect of our lives. In recent years, Knowledge Graphs technology as one of the artificial intelligence techniques has been widely used in the educational domain. However, there are few studies dedicating the construction of knowledge graphs for K-10 education in Australia, and most of the existing studies only focus on at the theory level, and little research shows practical pipeline steps to complete the complex flow of constructing the educational knowledge graph. Apart from that, most studies focused on concept entities and their relations but ignored the features of concept entities and the relations between learning knowledge points and required learning outcomes. To overcome these shortages and provide the data foundation for the development of downstream research and applications in this educational domain, the construction processes of building a knowledge graph for Australian K-10 education were analyzed at the theory level and implemented in a practical way in this research. We took the Year 9 science course as a typical data source example fed to the proposed method called K10EDU-RCF-KG to construct this educational knowledge graph and to enrich the features of entities in the knowledge graph. In the construction pipeline, a variety of techniques were employed to complete the building process. Firstly, the POI and OCR techniques were applied to convert Word and PDF format files into text, followed by developing an educational resources management platform where the machine-readable text could be stored in a relational database management system. Secondly, we designed an architecture framework as the guidance of the construction pipeline. According to this architecture, the educational ontology was initially designed, and a backend microservice was developed to process the entity extraction and relation extraction by NLP-NER and probabilistic association rule mining algorithms, respectively. We also adopted the NLP-POS technique to find out the neighbor adjectives related to entitles to enrich features of these concept entitles. In addition, a subject dictionary was introduced during the refinement process of the knowledge graph, which reduced the data noise rate of the knowledge graph entities. Furthermore, the connections between learning outcome entities and topic knowledge point entities were directly connected, which provides a clear and efficient way to identify what corresponding learning objectives are related to the learning unit. Finally, a set of REST APIs for querying this educational knowledge graph were developed.

FoR codes (2008)

0803 COMPUTER SOFTWARE, 0804 DATA FORMAT, 1301 EDUCATION SYSTEMS, 1302 CURRICULUM AND PEDAGOGY

Download

Share

COinS
 

Unless otherwise indicated, the views expressed in this thesis are those of the author and do not necessarily represent the views of the University of Wollongong.