Component-level embeddings make CourseGraph interpretable: learning outcomes contributed most to an example overlap decision¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? In an interpretable mapping example, the learning outcomes contributed the most to the overlap decision between Distributed Data Systems (TUD) and Distributed Systems (LU). → Arthur Nijdam 2026

Evidence¶

Arthur Nijdam 2026¶

Arthur Nijdam, Paul Stankovski Wagner, Sara Ramezanian. (2026). CourseGraph: Finding overlaps and differences in Computer Science courses across universities. https://arxiv.org/abs/2608.05910 doi:10.48550/arxiv.2608.05910

q2 · i? · design · r2

Case analysis of one CourseGraph mapping (Table II) shows cosine similarity of 0.81 for learning outcomes versus 0.70 for descriptions between the TUD and LU courses; the authors note the similarity "is significantly higher for the LOs than for the other course components".

"We see that the LOs contributed the most to this particular decision. Upon inspecting the data, this makes sense: The description of the course at TUD is a lot longer and contains a more marketing-focused tone."

Discussion¶

Related Claims¶

• In a worked example from real Erasmus+ decisions, some of CourseGraph's nearest-neighbor course mappings matched the program director's and others did not, one plausibly because it ignores differences in learning methods — related
• NCA scores of IIU input vectors order action types identically to the main-effects ordering (Spearman's ρ = 1) — related
• On the TU/e CS overlap dataset, supervised NLP classifiers (Random Forest, XGBoost) outperform thresholding and zero-shot LLM baselines at detecting course overlap — related