level: research
undergraduate computer science programs follow international guidelines that are updated about once a decade. a new human-in-the-loop pipeline measures how completely a program covers the current body of knowledge. it was tested on one accredited bachelor's program against the cs2013 and cs2023 standards. the pipeline treats the program and each guideline as structured text corpora. it first generates candidate matches between courses and knowledge units using semantic retrieval, then confirms them through human judgment under a clear coverage definition.
seven different retrieval methods were benchmarked. a reciprocal-rank-fusion ensemble performed best. a well-known long-context model did worse than a small sentence model. this shows that retriever choice matters for accurate alignment. the approach provides a reproducible way to track coverage over time, especially when guidelines are restructured. it helps programs identify gaps and overlaps in their curriculum relative to the latest standards.
the framework supports longitudinal analysis, letting institutions see how their coverage shifts from one guideline version to the next. this is useful for accreditation and continuous improvement. the method combines automated semantic matching with human oversight to ensure reliable results. it moves beyond simple keyword matching to capture deeper topical and competency alignment. the pipeline can be adapted to other disciplines and bodies of knowledge.
why it matters: it gives cs departments a data-driven way to align courses with evolving industry standards, ensuring graduates have relevant skills.