TITLE:
Ontology-Guided Semantic Retrieval in Virtual University Information Systems: An Empirical Benchmark Study
AUTHORS:
Mamadou Bakouan, Tiemoman Kone, Yao Konan
KEYWORDS:
Semantic Retrieval, Ontology-Based Retrieval, Information Retrieval, Virtual University, Sensitivity Analysis
JOURNAL NAME:
Open Journal of Applied Sciences,
Vol.16 No.9,
September
8,
2026
ABSTRACT: Semantic retrieval in virtual university information systems must address lexical variation and domain specific relationships. This study reports a controlled empirical evaluation of lexical, semantic, ontology-based, and combined retrieval configurations within a virtual-university benchmark. The benchmark contains 20 queries and 38 documents, with 70 judged query document pairs and 690 un judged pairs. Five configurations are evaluated using P5, R5, F1_5, MRR, and nDCG5, with query level analysis and multiplicity controlled statistical comparisons. The combined configurations achieve the highest recorded aggregate scores, including nDCG5 = 0.9894. The largest recorded descriptive difference relative to ontology only retrieval is ΔnDCG5 = 0.07308; the corresponding recorded Holmadjusted p-value is 0.05388 and therefore does not meet the predefined 0.05 significance threshold. Sensitivity analysis identifies a local performance plateau for the tested ontology-dominant interval α = 0 and β = 0.05 - 0.30. The findings provide benchmark-specific descriptive and suggestive evidence for combining semantic and explicit domain knowledge signals, while indicating the need for larger benchmarks and independently reproducible query level statistical analyses.