Gelal, Nirmal2025-10-282025https://hdl.handle.net/2097/45412The selection of diverse, thematically aligned literature is a significant challenge for high school English teachers due to limited time and resources. This study presents T-TExTS, a knowledge graph-based recommendation system designed to address this problem by scaffolding educators in their text selection process. We constructed a domain-specific ontology using KNowledge Acquisition and Representation Methodology (KNARM), which was then instantiated into a knowledge graph to power the recommendation engine. Our core contribution is a comparative analysis of two graph embedding paradigms: shallow methods (DeepWalk, Biased Random Walk, and Hybrid model) and a deep method (Relational Graph Convolutional Network, R-GCN). The models were evaluated on both link prediction and recommendation ranking tasks. While the shallow DeepWalk model achieved the highest AUC for link prediction (0.9739), the deep R-GCN model proved superior for the primary tasks of recommendation ranking, outperforming other models on metrics such as Hits@10, MRR, and nDCG@10. This finding supports our hypothesis that deep embedding approaches, by capturing richer relational semantics, are better suited for recommendation tasks on knowledge-augmented datasets. The results demonstrate that T-TExTS provides an effective, ontology-driven solution to assist educators in making more informed and inclusive curricular decisions.en-USKnowledge GraphSemantic WebRepresentational LearningEmbedding MethodsRecommendationThesis