Year: 2026
Type: dissertation
Abstract: Knowledge Graphs (KGs) play a key role in organizing structured data. However, their incompleteness limits applications in education and intelligent tutoring. This thesis investigates neural network methods for building, completing, and reasoning over KGs in educational settings. Four studies are presented. First, it evaluates Large Language Models (LLMs), such as GPT-4, for constructing and expanding Educational Knowledge Graphs, achieving high semantic similarity (0.875) with instructor-authored content. Second, it explores spiking neural network (SNN) architectures for interpretable, neuromorphic relational inference in KGs. Third, it develops an SNN-based KG embedding framework with learnable synaptic delays for energy-efficient prerequisite reasoning in intelligent tutoring systems. Finally, it compares Cache-Augmented Generation (CAG), Retrieval-Augmented Generation (RAG), and long-context prompting for generating curriculum knowledge graphs from course materials across 36 courses, revealing trade-offs between graph size and local deployability.
Source: AUSpace (Athabasca University)
Author: Jhajj, Gaganpreet
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: green
OpenAlex ID: W7222376170