RAGEAR: Retrieval-Augmented Graph-Enhanced Academic Recommender

RAGEAR is a neurosymbolic recommender system that combines dense retrieval of lecture transcripts with knowledge graphs to improve academic course recommendations. The system demonstrates that fine-grained instructional content outperforms metadata-only approaches, with a novel graph-aware aggregation function that effectively propagates evidence from transcript chunks to course-level rankings.

RAGEAR addresses a practical gap in academic advising by moving beyond traditional metadata-based course recommendation systems. The architecture integrates two complementary approaches: dense retrieval captures semantic relevance from detailed lecture content, while symbolic knowledge graphs enforce structured constraints like prerequisites and credit requirements. This hybrid neurosymbolic design reflects a broader industry trend toward combining neural flexibility with symbolic reasoning for interpretability and constraint satisfaction.

The technical innovation lies in the graph-aware aggregation function, which doesn't simply sum retrieval scores but accounts for three factors: how much retrieved content belongs to each course, the ranking strength of relevant chunks, and evidence distribution across lessons. This prevents courses from gaining unfair advantages through a single highly-relevant lecture while ignoring weak coverage elsewhere. The evaluation methodology—combining human assessment with LLM-based relevance scoring across 152 queries—represents solid empirical validation beyond typical benchmark testing.

For educational technology stakeholders, RAGEAR demonstrates measurable returns from investing in detailed content indexing and sophisticated retrieval pipelines. Universities and EdTech platforms using course recommendation systems can expect quality improvements from transcript-based approaches over simple metadata filtering. The work validates that fine-grained content analysis creates competitive advantages in personalization, potentially enabling better course matching and improved student outcomes. The research suggests future systems should prioritize rich content analysis combined with structured knowledge representation rather than relying on surface-level course attributes.

• →RAGEAR combines dense transcript retrieval with knowledge graphs for more nuanced course recommendations than metadata-only systems
• →The graph-aware aggregation function uniquely propagates chunk-level evidence to course scores, improving ranking quality especially for top results
• →Fine-grained instructional content analysis outperforms traditional metadata-based approaches in academic recommendation tasks
• →The neurosymbolic design balances neural flexibility for semantic matching with symbolic reasoning for constraint enforcement
• →Human and LLM-based evaluation confirms measurable improvements over transcript-based baseline approaches