LIT-GRAPH: Evaluating Deep vs. Shallow Graph Embeddings for High-Quality Text Recommendation in Domain-Specific Knowledge Graphs

TL;DR

The study addresses curriculum stagnation in high school English by applying a domain-specific Knowledge Graph (KG) to power literature recommendations. It compares shallow embedding methods (DeepWalk, Biased Random Walk, Hybrid) with a deep Relational Graph Convolutional Network (R-GCN) on a KNARM-derived ontology, evaluating both link prediction and ranking quality. DeepWalk delivers strong link-prediction performance ($AUC=0.9737$), while R-GCN excels in semantic ranking metrics (e.g., $Hits@10=0.7368$, $nDCG@10=0.4985$), highlighting the value of relation-aware message passing for pedagogy-focused recommendations. The findings imply that deep relational embeddings better capture domain-specific semantics, guiding educators toward high-quality, pedagogically aligned texts; future work includes scalable KG construction with LLMs and GraphRAG-based interactive explanations.

Abstract

This study presents LIT-GRAPH (Literature Graph for Recommendation and Pedagogical Heuristics), a novel knowledge graph-based recommendation system designed to scaffold high school English teachers in selecting diverse, pedagogically aligned instructional literature. The system is built upon an ontology for English literature, addressing the challenge of curriculum stagnation, where we compare four graph embedding paradigms: DeepWalk, Biased Random Walk (BRW), Hybrid (concatenated DeepWalk and BRW vectors), and the deep model Relational Graph Convolutional Network (R-GCN). Results reveal a critical divergence: while shallow models excelled in structural link prediction, R-GCN dominated semantic ranking. By leveraging relation-specific message passing, the deep model prioritizes pedagogical relevance over raw connectivity, resulting in superior, high-quality, domain-specific recommendations.

Figures (1)

• Figure 1: Workflow of the English Literature Text Recommender System. Note: The Hybrid model is the concatenation of vector outputs from DeepWalk (DW) and Biased Random Walk (BRW). R-GCN: Relational Graph Convolutional Network.