Column
Your Training Curriculum and Job Requirements Are Probably Out of Sync — NLP Can Now Measure the Gap
An NLP pipeline that auto-extracts skills from curricula and job postings, maps them to the ESCO standard classification, and quantifies the gap by category is now available as a research framework. The structure translates directly to corporate L&D and hiring strategy.
Hi, I’m Keito Inoshita from Affectosphere Group.
Is your internal training program actually building the skills your organization needs right now?
This is a question many L&D teams can’t answer with data. The program was designed at a point in time when the skill requirements were different. The business has shifted, hiring requirements have evolved, and somewhere in the gap between then and now, the training content and the actual need drifted apart. But because there’s no quantitative way to measure the drift, the priority to update it stays low.
A paper published on arXiv in June 2026 (Sherzod Turaev, Mary John, Mamoun Awad, Nazar Zaki, Khaled Shuaib et al., arXiv:2606.01982) was designed for higher education reform, but the pipeline it builds is directly applicable to the corporate L&D context. The framework automatically extracts skills from curricula and job postings using LLMs, maps them to the ESCO EU skill classification standard, and quantifies supply-demand gaps across multiple dimensions.
3 Points for Today
- Value: You can now quantify the gap between what your training delivers (supply) and what your roles require (demand), using a standardized classification that enables cross-category comparison.
- How it works: A 4-stage pipeline — LLM extraction, ESCO mapping, gap quantification — running on documents you already have.
- Business application: Training investment prioritization, build-vs-buy hiring decisions, and L&D strategy backed by data rather than intuition.
① What the Research Solved
The problem the research addresses: the two sides of the education-employment system speak different languages. Curricula describe learning outcomes in pedagogical terms. Job postings describe requirements in operational terms. Without a shared vocabulary, comparing what’s being taught against what’s being required is intractable.
The four-stage solution:
First, LLMs (GPT-4o, Gemini, etc.) extract competencies from both document types using schema-constrained prompts — output structure is enforced, achieving 100% schema compliance and full document coverage.
Second, extracted skills are mapped to ESCO (European Skills, Competences, Qualifications and Occupations), an EU-managed standard classification for skills across all industries. Using a public standard rather than an internal taxonomy enables cross-organizational comparison.
Third, the mapped results quantify both supply coverage (what the curriculum delivers) and demand frequency (what jobs require) across ESCO categories.
Fourth, the comparison is translated into a multi-dimensional gap score by category.
Applied to 85 computer science courses and 30 job postings at a UAE university, the framework identified a 25% gap in generic skills and a 13.8% gap in algorithm theory. The AI-specific skill gap was relatively small. Inter-rater reliability (Cohen’s kappa) reached 0.79.
② How Companies Can Use This
The same pipeline applies directly to corporate training.
Automatically visualize gaps between training curriculum and job requirements
L&D manages course catalogs and training syllabi. HR manages job descriptions. Run both through the same pipeline, mapped to ESCO, and you get a quantified answer to: what percentage of current job requirements does existing training cover? Which skill categories have the largest gaps?
This turns a feeling of misalignment into a number. And a number can go into an executive deck.
Set training investment priorities with data behind them
Once gaps are quantified by category, the prioritization conversation changes. “Our generic skills gap is 25%” provides a grounded basis for proposing expansion of communication, problem-solving, or project management training. “Our AI skills gap is small relative to other areas” supports a decision not to allocate new budget there this cycle.
Investment decisions that previously rested on program manager intuition or anecdotal feedback now have an objective baseline.
Make build-vs-hire decisions more precise
For any specific skill cluster: if the gap analysis shows the training program already covers 70% of what the role requires, the case for internal development is stronger. If it covers 20%, the case for external hiring is stronger.
The boundary between “develop internally” and “hire for” has historically been drawn by gut feeling and market availability. This pipeline adds a quantitative layer to that decision.
③ What Implementation Requires
Running this in a corporate context requires: document access (existing syllabi, JDs, skills frameworks); LLM API access (GPT-4o equivalent); and ESCO API access (free, publicly available).
A practical starting point: run a PoC on one department — 10-20 training courses and 10-20 current JDs — before committing to a broader build.
Important caveat: the paper’s results are from English-language text in a computer science domain. Reliability in other languages or industry domains requires separate validation. For Japanese-language environments, ESCO’s Japanese coverage and LLM performance on Japanese technical vocabulary are worth testing before drawing conclusions.
The 0.79 kappa score is a strong indicator of reliability within the tested conditions. The honest framing for corporate use is: it’s a validated approach in a specific context, and your context will require its own validation pass.
Turning “Something Feels Off” Into a Number
What L&D practitioners struggle with most is not identifying that training is misaligned — they often sense it clearly — but demonstrating the degree of misalignment in terms that justify a budget reallocation.
That’s what this pipeline contributes: not a new training theory, but a measurement instrument that turns a qualitative perception of drift into a category-level gap score.
The implementation cost to run this yourself is real. But the knowledge that this level of analysis is now technically feasible is itself useful for anyone thinking about the next generation of L&D infrastructure.
That’s all for today!
References
- Sherzod Turaev, Mary John, Mamoun Awad, Nazar Zaki, Khaled Shuaib (2026). An NLP-Driven Framework for Curriculum-Labor Market Alignment: Schema-Constrained LLM Extraction, ESCO-Anchored Semantic Matching, and Multi-Dimensional Gap Quantification. arXiv preprint.
* This article was written in part with AI assistance and may contain inaccuracies.