Table 1.

A summary of selected KG construction approaches for adaptive and personalised learning.

Ref.	KG Specific Purpose	Construction Algorithm(s)	Type of KB	KG Resource(s)	#Entites (e)/#Relations (r)	Evaluation Criteria	Limitation(s)
[58]	Learning assessment and recommendation	Bootstrapping construction strategy and BERT-BiLSTM-CRF	Schema-free	Subject teaching resources, Baidu Encyclopedia, and DBPedia	#e: 2202 #r: 3122	P, R, F1 measure, and case study	•Restricted presentation on the benefit of the constructed KG for student learning,
[33]	Learning Resource Recommendation	Ontology construction, Weighted Fusion Method	Schema-based	Learning resources	N/A	F1-score comparison, Efficiency analysis	•Limited focus on discrete mathematics, •Assumed optimal alpha, within a specific range, •Small-scale dataset
[34]	Intelligent Tutoring System for Math Education	DL-based Grading Model, STACK-based Grading Model,	N/A	Learning resources, Learner profiles, Grading models, Instructional concepts, Knowledge states, Learning interactions	N/A	Quadratic weighted kappa value, F1-score, Accuracy of grading models	•Limited focus on discrete mathematics, •Third-party reliance on STACK, •Small-scale initial deployment.
[35]	personalised learning path recommendation	multi-dimensional KG frameworks, attention mechanisms, and activation theory for path generation	N/A	Educational resources	N/A	Accuracy, effectiveness, and quality of adaptive learning services.	•Refining frameworks for various learning scenarios, •Scalability in path generation and •Addressing limitations in automatic cognitive perception within the KG.
[36]	Students' clustering and course recommendation	knowledge network, machine-learning	Schema-free	Student profiles, course data, and features extracted from textual data	#e: 675 #r: 1033	P, R, Accuracy, F1_Score RMSE and MAP	•Data quality, •Scalability, •Biases in recommendations, •Challenges in accurately capturing students' preferences and learning behaviours.
[37]	learning resources and guiding recommendations	N/A	Hybrid-based	Educational content	N/A	N/A	•Lack of discussion on KG construction. •Lack of discussion on the evaluation criteria
[38]	Visual representation of learning paths	Concept maps	Hybrid-based	knowledge units	N/A	Case Study	•Challenges related to the accuracy of cognitive reasoning, •Lack of rigorous evaluation metrics.
[39]	Adaptive learning experiences for students	An improved version of the FP-growth algorithm	Schema-based	Students' searches within the online learning system	N/A	Students' satisfaction via surveys	•Poor evaluation metrics, •Limited discussion on the collected entities and relationships.
[40]	Adaptive E-learning for Adult Learners in Open Education	Manual extraction of entities and relationship	Schema-based	Learning resources of the course “Principle and Application of Database System”	N/A	Subjective evaluation of 30 learners who participated in an online course.	•Potential difficulties in managing and updating the knowledge graph as the course content evolves. •Limited discussion on mechanisms used to construct the KG as well as the size of the resultant graph.
[48]	Identify students at risk of failing a course and provide personalised interventions.	Ontology mapping	Hybrid	Courses offered by the College of Information Technology at UAEU between 2016 and 2021.	N/A	P, R, F1-score, and Accuracy	•Limited dataset size, •Lack of statistics with regards to the size of resultant KG, •Lack of subjective evaluation and application of downstream task.
[49]	Development of an interpretable early warning recommendation mechanism for learning behavior.	DNNs	Hybrid	AI-enabled online learning platform	#e: 1204	AUC, RI, F1 score, and Multi-task Learning Gain (MTL-Gain).	•There are in-depth logical designs and topology verifications for concept classes of learning content, •A need to improve the dependability of feedback and early warning accuracy.