Researchers have developed and demonstrated an open, artificial intelligence-enabled methodology called the Curriculum Analytics Framework to evaluate how closely university degree programmes align with external labour market skill demands. Demonstrated through an empirical investigation of a bachelor programme in economics and business at the Stockholm School of Economics in Riga, Latvia, the framework transforms unstructured syllabus documents and tens of thousands of online job postings into standardised, comparable skill representations. Rather than relying on intuition, subjective surveys, or commercial black-box platforms, the newly published framework introduces a reproducible analytical pipeline that pairs large language model processing with official European occupational taxonomies to diagnose educational strengths, curricular blind spots, and structural differences in skill prioritisation across multiple graduate employment pathways.
The study addresses a longstanding structural friction in higher education governance, where academic institutions face mounting external pressure to adapt their learning programmes to rapid technological disruption, digital transformation, and demographic shifts. Historically, universities have evaluated their educational offerings through periodic internal reviews, accreditation procedures, alumni surveys, and advisory consultations. While these traditional mechanisms provide vital qualitative context, they remain periodic, small in scale, and largely disconnected from real-time empirical signals emerging across broader employment sectors. Although recent advances in educational data mining and learning analytics have expanded data utilisation across universities, such applications have concentrated heavily on student-facing metrics such as exam performance, engagement monitoring, retention, and student progression, leaving curriculum-level analytics comparatively underdeveloped.
To bridge the operational gap between institutional educational provision and external employer demand, the authors designed the Curriculum Analytics Framework around a structured five-step sequence. The workflow begins by compiling and standardising observational labour market data from scraped job advertisements through an AI-assisted classification pipeline. Second, the system extracts skill profiles from university course outlines and converts them into structured supply vectors using the same taxonomic classifications. Third, rather than comparing a specialised academic programme against an undifferentiated national labour market, the framework defines empirical occupational benchmarks based on observed graduate placement pathways. Fourth, the architecture calculates mathematically comparable measures of labour market skill demand and curricular skill supply. Finally, the framework evaluates alignment across three complementary analytical metrics: the Weighted Mean Gap, Spearman rank correlation, and Importance-Weighted Coverage.
Constructing the empirical labour market dataset required scraping and standardising 43,522 job advertisements collected from five major recruitment platforms in Latvia between June 2025 and February 2026. Because raw recruitment listings lack uniform occupational tags and contain highly irregular terminology, the authors implemented an AI-assisted classification pipeline powered by Anthropic’s Claude Opus 4.6. Operating under strict, researcher-supervised rules and validation protocols, the model classified each advertisement into the European Skills, Competences, Qualifications and Occupations taxonomy and extracted specific skill requirements. In a subsequent hold-out validation involving 162 randomly sampled advertisements independently blind-coded by an external expert, the automated pipeline achieved an exact occupational agreement of 90.7 percent and reached 93.8 percent agreement at the four-digit occupational-group tier, confirming robust classification consistency.
Standardising raw skill phrases extracted from both job advertisements and course syllabi proved essential because employers and academics routinely use disparate wording to describe identical capabilities. A competence such as Python programming might appear as several distinct text strings, including coding in Python or Python software development. To resolve this fragmentation without losing analytical clarity, the authors mapped all extracted competences to the ESCO Level 3 grouping layer, which comprises 302 distinct skill categories. This intermediate layer was selected because it successfully balances contextual granularity against statistical tractability, avoiding both the oversimplification of broader categories and the extreme dispersion of the full leaf-level taxonomy, which encompasses more than 13,000 granular concepts.
On the educational supply side, the investigation analysed the core curriculum of the Bachelor of Science in Economics and Business at SSE Riga, consisting of 29 mandatory core courses totalling 130 European Credit Transfer and Accumulation System credits. By extracting competences directly from complete course outlines—including intended learning outcomes, lecture programmes, and assessment tasks—the language model generated an initial curriculum inventory that mapped to 63 ESCO Level 3 skill categories. To establish an empirically grounded demand benchmark, the researchers analysed 85 observed graduate employment placements gathered from the institution’s 2025 placement monitoring survey. These placements were grouped into nine professional clusters, such as Business Analysis and Consulting, Finance and Accounting, and Marketing and Communications, which were linked to 4,629 relevant junior- and mid-level job postings.
To evaluate alignment accurately, the researchers devised distinct quantitative scaling mechanisms to overcome the fundamental structural differences between large vacancy collections and compact curricular course sets. On the demand side, simple word frequencies were rejected because they disproportionately highlight ubiquitous, non-distinctive capabilities such as basic teamwork or general communication. Instead, the authors adapted Revealed Comparative Advantage metrics to identify competencies that appear with disproportionate frequency within specific occupational clusters relative to the broader labour market. These scores underwent logarithmic transformation and maximum normalisation within each occupational group. On the supply side, curriculum competencies were weighted by the relative credit value of each course, reflecting instructional emphasis, and normalised against cluster-specific skill demands.
The framework avoids aggregating curriculum alignment into an oversimplified single index, relying instead on three distinct, complementary metrics to diagnose educational positioning. The Weighted Mean Gap measures the numerical direction and magnitude of emphasis differences between curricular delivery and employer requests, highlighting whether a given competence receives higher or lower relative instructional weight. Spearman rank correlation evaluates whether overlapping skills share a similar hierarchy of prioritisation between faculty syllabi and market postings. Finally, Importance-Weighted Coverage calculates the proportional share of total labour market skill importance represented within the curriculum, measuring whether high-priority capabilities appear anywhere in the degree programme regardless of their exact instructional credit weight.
Applying these metrics across the nine graduate career clusters revealed substantial variations in curriculum alignment. Business Analysis and Consulting exhibited the strongest overall alignment, recording the highest Importance-Weighted Coverage score of 0.727 alongside the highest positive Spearman rank correlation. Core courses placed pronounced instructional weight on analysing data, developing solutions, and quantitative evaluation, which matched employer priorities closely, although positive gaps emerged in applied capabilities such as IT systems design. In Finance and Accounting, which represented the destination for 48.2 percent of tracked graduates, the degree programme demonstrated strong analytical foundations and positive rank alignment of 0.20 across 49 overlapping competencies, while exhibiting smaller gaps in specialised tasks such as budget management and routine financial reporting.
Conversely, the multidimensional metrics revealed important structural misalignments in pathways such as Human Resources and People Management. While this occupational cluster achieved relatively high skill coverage, it generated a negative Spearman rank correlation of minus 0.34 across 47 overlapping skills. This divergence occurred because the academic programme prioritised broad analytical reasoning, whereas the labour market demanded specialised competencies such as candidate interviewing, personnel administration, and performance evaluation. A similar negative rank correlation and lower coverage score of 0.479 appeared in Real Estate and Property, while Administration and Support displayed positive skill gaps reflecting the programme’s analytical orientation over operational administrative execution.
To verify the stability of these findings, the research team conducted three rigorous sensitivity and robustness checks across the dataset. A non-parametric bootstrap procedure resampling job postings across 500 iterations confirmed narrow 95 percent confidence intervals across most clusters, demonstrating that results were not artefacts of random sampling fluctuations. A leave-one-skill-out jackknife sensitivity test confirmed that the sign of the Spearman rank correlation remained completely stable in eight of the nine clusters, shifting only in the small Public Administration cluster. Furthermore, re-running the complete analytical pipeline after eliminating 7,466 advertisements that contained five or fewer mapped skills produced virtually identical cluster rankings, with Jaccard similarity indices for top-demanded skills ranging from 0.88 to 1.00.
The authors emphasise that the Curriculum Analytics Framework is designed strictly as a diagnostic decision-support system rather than an automated mandate for institutional syllabus revision. Curricular data reflect intended, documented learning designs rather than informal classroom interactions or individual graduate mastery, while online postings capture publicly advertised needs rather than informal hiring networks. Consequently, observed gaps should prompt deliberative dialogue among academic leadership, instructors, and industry stakeholders rather than hasty course restructuring. Foundational academic concepts may purposefully diverge from immediate entry-level job notices, and narrower technical skills may be more effectively delivered through specialized elective tracks, internships, or practical modules rather than mandatory core coursework.
Subject of Research: AI-enabled curriculum analytics framework for university and labour market skill alignment
Article Title: AI-enabled curriculum analytics framework for aligning higher education with labour market skill demand
Article References: Varpina, Z., Kuzmina, A., Kraupsa, K., Bileskalne, L., & Culkstena, Z. (2026). AI-enabled curriculum analytics framework for aligning higher education with labour market skill demand. Discover Artificial Intelligence, 6(1), Article 1674. https://doi.org/10.1007/s44163-026-02627-9
Image Credits: AI Generated
DOI: 10.1007/s44163-026-02627-9
Keywords: Curriculum analytics, Labour market skill demand, Higher education, Artificial intelligence, Natural language processing, Large language models, ESCO taxonomy, Skills intelligence, Revealed comparative advantage, Educational data mining, Job vacancy data, Curriculum alignment, Stockholm School of Economics in Riga, Learning analytics, Workforce development, Occupational classification