An NLP-Driven Framework for Curriculum-Labor Market Alignment: Schema-Constrained LLM Extraction, ESCO-Anchored Semantic Matching, and Multi-Dimensional Gap Quantification

Abstract: Schema-constrained information extraction from diverse educational and labor-market corpora remains an open challenge in natural language processing because existing pipelines rely primarily on lexical-surface methods that cannot recover implicit competencies, lack grounding in shared taxonomies, and provide no formal measures of extraction reliability or document-level completeness. To address these limitations, this paper proposes a four-stage NLP framework that combines (i) schema-constrained prompting of a two-model frontier-LLM ensemble against a JSON Schema-enforced seven-slot competency formalism, (ii) Sentence-BERT (SBERT) alignment of the extracted records against an eleven-domain ESCO v1.2.1 controlled vocabulary, (iii) a two-tier adjudication protocol that resolves inter-model disagreements, and (iv) a verification mechanism that combines per-slot Cohen's kappa, schema conformance, and document-level completeness audits. The framework is instantiated for a critical application in higher-education quality assurance, namely curriculum-labor market alignment for the ABET-accredited BSc Computer Science program at the United Arab Emirates University. The pipeline extracts 400 competency records from the 85-course 2025-2026 study plan and aligns them, under a five-scope analysis ranging from the computing core to a probability-weighted student trajectory, with 30 job postings (483 requirement clauses) at an SBERT cosine threshold of 0.50. The extractor achieves Cohen's kappa of 0.79 on the skill slot, with 100% schema conformance and 100% document-level completeness. The alignment surfaces interpretable supply-demand gaps of 25.0% in general and transversal skills, 13.8% in algorithms and computational theory, and 12.2% in software engineering and project management, with a near-zero 1.8% gap in artificial intelligence and data science despite 38.6% supply coverage.

• The paper introduces a seven-slot schema and dual-LLM verification to reliably extract competencies from both curricula and labor market data.
• It applies ESCO anchoring, Bloom’s taxonomy, and SBERT semantic matching to quantify supply-demand gaps and cognitive depth differentials.
• The study offers actionable curriculum redesign recommendations and demonstrates the scalability of data-driven curriculum intelligence.

Schema-Constrained NLP for Curriculum-Labor Market Alignment: Methodology and Empirical Results

Problem Formalization and Previous Literature

Structured extraction of competencies from educational and labor-market corpora remains underdeveloped despite the pedagogical and economic urgency for evidence-based curriculum redesign. Prior computational approaches—TF-IDF weighting, topic modeling, shallow classifiers, and NER—either operate at the lexical surface or rely on ad hoc taxonomies, precluding reliable identification of implicit competencies and inhibiting any systematic cross-study comparison. Furthermore, extraction reliability is rarely quantified, leaving the trustworthiness of mismatches unvalidated.

This paper formally operationalizes curriculum-labor market alignment as a multi-dimensional matching problem. Competency extraction $E$ maps documents to structured records in a rigorously defined seven-slot formalism: $$(\text{label}, \text{domain}, \text{knowledge}, \text{skill}, \text{level}, \text{context}, \text{evidence})$$, using ESCO domains and Bloom's taxonomy for semantic anchoring and cognitive-level assignment. Alignment metrics encompass supply-demand coverage, Bloom's-level depth differential, and temporal emergence lag.

Framework Architecture

The proposed pipeline consists of four stages:

1. Corpus Construction: Parallel assembly of supply-side curricular documents (syllabi, CLOs, LOAMS, catalog course descriptions) and demand-side job advertisements. Rigorous preprocessing, including language detection, deduplication, and sentence segmentation, is executed with no stemming or lemmatization to preserve context for LLM extraction.
2. Schema-Constrained Competency Extraction: Schema-enforced prompting is applied to a two-model frontier LLM ensemble (OpenAI GPT-5.4 and Anthropic Claude Opus 4.6) for each document segment, targeting a JSON Schema formalism. Each LLM extracts a competency record per CLO, with slot values grounded in ESCO domains and Bloom's taxonomy. Multi-model verification is performed: agreement leads to automatic acceptance; disagreement is resolved through human adjudication.
3. Semantic Alignment and Gap Quantification: Extracted competencies are mapped to an ESCO-anchored embedding space using sentence-transformer (all-MiniLM-L6-v2). Gap quantification comprises SBERT cosine similarity-based coverage against the ESCO taxonomy, depth differential in Bloom's-level assignment, and a probability-weighted supply-side analysis modeling student elective trajectories.
4. Visualization and Decision Support: Results are presented as course-level heatmaps and program-level dashboards. Interpretability is enhanced by generating natural language summaries for each gap and actionable recommendations for program governance.

Empirical Evaluation and Results

Corpus

The pipeline was instantiated on the ABET-accredited BSc Computer Science program at UAEU using 397 CLOs from 85 courses (32 computing core, 8 supporting math/science, 45 gen-ed). Thirty regionally representative job postings (filtered for ISCO-08 groups 25, 35) provided 483 requirement clauses for demand-side alignment.

Extraction Verification

Manual audit of 50 records yielded Cohen's $\kappa = 0.79$ for the skill slot (substantial agreement), $\kappa = 0.55$ for knowledge, $\kappa = 0.43$ for domain, and 100% schema/document completeness. Domain-slot ambiguity and collapsed knowledge/skill slots were identified as open problems for prompt refinement.

Coverage Gaps

Supply-demand gaps were quantified at the SBERT cosine threshold of 0.50:

• General and Transversal Skills: 25.0% supply-demand gap (supply coverage 25.0%, demand coverage 41.7%), spanning critical thinking, leadership, and time management.
• Algorithms and Computational Theory: 13.8% gap, predominantly tool-specific proficiencies (ASP.NET, C#, C++, PHP, Python, TypeScript) not matched by the curriculum at ESCO granularity.
• Software Engineering and Project Management: 12.2% gap, reflecting demand for Agile, technical documentation, and leadership not fully covered by existing courses.
• Artificial Intelligence and Data Science: Near-zero gap (1.8%), with AI supply coverage (38.6%) far exceeding demand, contradicting common assumptions of AI skill shortfall.
• Cybersecurity and Ethics: Modest coverage gap (6.2%) but lowest mean Bloom's level ($\mu = 2.50$) and no competencies at Create-level.
• Other domains (web/mobile dev, systems/infrastructure, programming/software dev, HCI/design) exhibited gaps $\leq 7\%$.

Depth Differential

The mean Bloom's level for computing-core records was $\mu = 3.17$ (Apply), with substantial domain-level variation. HCI and Design ($\mu = 3.85$), Algorithms ($\mu = 3.83$), Programming ($$(\text{label}, \text{domain}, \text{knowledge}, \text{skill}, \text{level}, \text{context}, \text{evidence})$$0) had higher cognitive depth, while Cybersecurity/Ethics ($$(\text{label}, \text{domain}, \text{knowledge}, \text{skill}, \text{level}, \text{context}, \text{evidence})$$1) and General/Transversal ($$(\text{label}, \text{domain}, \text{knowledge}, \text{skill}, \text{level}, \text{context}, \text{evidence})$$2) lagged.

Scope Sensitivity

Analysis across five supply-side scopes (computing core, disciplinary, full program, deterministic student path, probability-weighted path) showed that naive aggregation of gen-ed electives artificially reduced transversal gaps. The probability-weighted path, accounting for actual student trajectories, preserved the 25.0% gap, confirming the diagnostic fidelity of scope-weighting.

Implications and Theoretical Contributions

Methodological Advancement

• Seven-Slot Formalism: Enables decomposed, evaluable representations and supports depth-gap and temporal lag analysis unavailable in previous studies.
• Two-Model Adjudication Protocol: Reduces hallucination, quantifies reliability, and enables conservative governance stratification.
• ESCO-Anchored Taxonomy: Normalizes semantic comparisons across registers and institutional conventions, mitigating false negatives/positives inherent to lexical methods.
• Probability-Weighted Scope: Realistically models student achievement, preventing overestimation of generic competencies via catalog aggregates.

Practical Impact

The evidence produced addresses critical domains for intervention, moving program committees from survey-driven review toward continuous, data-driven curriculum intelligence. Recommendations include:

• Introduction of dedicated professional-competency courses for transversal skills.
• Curricular restructuring for software engineering, emphasizing project management and process methodologies.
• Course redesign in cybersecurity, prioritizing procedural over conceptual learning.
• Modular enrichment and agile revision for tool- and technology-specific gaps.

Limitations and Future Work

Current pilot demand corpus limits statistical inferences for the regional labor market. Further scaling is underway with 3,200 postings. Verification sample size and domain-slot ambiguity warrant expanded audit and robustness checks. Prompt refinement is required to address knowledge/skill collapse and context-slot underspecification.

Longitudinal deployment is planned to track temporal lag and evaluate the impact of interventions. Replication across multiple institutions and disciplines is needed to validate generalizability.

Conclusion

This work presents an integrated, schema-constrained NLP-driven framework for curriculum-labor market alignment, validated via empirical instantiation and rigorous verification (2606.01982). The approach advances methodological rigor, diagnostic fidelity, and actionable governance insight, demonstrating substantial progress toward automated, evidence-based curriculum intelligence that is replicable and generalizable across institutional contexts. The principal numerical findings inform both immediate curricular interventions and future comparative studies in education analytics and applied NLP.