---
title: "Measuring Digital Labour Market Transitions with a Digital"
description: "First, embedding retrieval proposes ten candidate occupations. 14 references. The paper tackles a measurement gap."
authors: ["Sadegh Shahmohammadi", "Xavier Pinho", "Mairi Bowdler", "Suhendan Adiguzel-van Zoelen", "Joost van Genabeek"]
year: 2026
venue: "arXiv:2608.24222"
source: "https://arxiv.org/abs/2608.24222"
doi: "10.48550/arxiv.2608.24222"
arxiv: "2608.24222"
licence: "CC BY-NC-SA 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by-nc-sa/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/shahmohammadi-2026-measuring-digital-labour-market-transitions/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/shahmohammadi-2026-measuring-digital-labour-market-transitions/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/36/) / Measuring Digital Labour Market Transitions with a Digital Semantic Score: An AI-Based Methodology Applied to the Dutch Labour Market

# Measuring Digital Labour Market Transitions with a Digital Semantic Score: An AI-Based Methodology Applied to the Dutch Labour Market

**Authors:** Sadegh Shahmohammadi, Xavier Pinho, Mairi Bowdler, Suhendan Adiguzel-van Zoelen and Joost van Genabeek · **Year:** 2026 · **Venue:** arXiv:2608.24222 · **Licence:** [CC BY-NC-SA 4.0 (arXiv)](http://creativecommons.org/licenses/by-nc-sa/4.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2608.24222) · [PDF](https://arxiv.org/pdf/2608.24222) · [DOI 10.48550/arxiv.2608.24222](https://doi.org/10.48550/arxiv.2608.24222) · [arXiv 2608.24222](https://arxiv.org/abs/2608.24222) · [OpenAlex](https://openalex.org/W7213863260)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper tackles a measurement gap. Standard labour statistics and keyword counts are too coarse to show how digitalisation changes the wording of jobs, the paths people take between occupations, and the skills they claim. The TNO authors use Revelio Labs profile data, largely derived from LinkedIn, and restrict it to Dutch job histories. They map free-text job titles to ESCO occupations in two steps. First, embedding retrieval proposes ten candidate occupations. Then an LLM picks one of them or declines to choose. The codes are then collapsed to four-digit ISCO-08. The paper also introduces a Digital Semantic Score, which rates a job title or skill by how much closer its embedding lies to a set of digital anchor concepts than to a non-digital reference set. The analysis finds that digitalisation is uneven. It is strongest in managerial, professional and ICT work, but is spreading into hybrid business, marketing and automation roles. Movement toward digital work depends on the occupation a person starts from. The resulting framework is scalable and can feed reskilling and skills-mismatch policy.

## Key points
- The data come from Revelio Labs and cover more than one billion profiles, 20 million companies and over 3,000 reported skills. The study uses only the Netherlands subset.
- For ESCO matching, OpenAI text-embedding-3-large (3,072 dimensions) embeds both job descriptions and ESCO entries. ESCO entries are built from preferred, alternative and hidden labels plus the occupation description. FAISS retrieves the top-10 candidates, and an LLM selects one with a structured prompt. When the LLM's confidence is low, no code is assigned.
- ESCO codes are reduced to four-digit ISCO-08, and the major group is taken from the first digit. Job-to-job transitions are built from consecutive records within each person. They are summarised as row-normalised shares and are explicitly treated as descriptive, not population-weighted.
- The Digital Semantic Score relies on text-embedding-3-small and compares each item, via cosine similarity, against openly specified sets of digital and non-digital anchors. It involves neither keyword matching nor a supervised classifier.
- The authors state several limitations:
  - The LinkedIn population over-represents white-collar workers.
  - Skills are attached to the whole profile, not to individual jobs, so skill change cannot be tracked over a career.
  - Skills are self-reported and therefore noisy.

## Interesting ideas
- Pairing high-recall embedding retrieval with an LLM that chooses only among the shortlist, and may decline, keeps the label space closed while gaining contextual disambiguation.
- Representing each taxonomy entry by concatenating its preferred, alternative and hidden labels with its gloss makes the entry far easier to match against messy surface text.
- Scoring an item by its similarity to a positive anchor set relative to a contrasting reference set yields a graded, inspectable property without training a classifier. Because the anchors are listed openly, anyone can audit or change them.
- Letting the classifier abstain under low confidence trades coverage for precision and keeps unreliable labels out of later analyses.
- The paper is careful to say what each measure cannot show, for example that profile-level skills say nothing about a particular job. This guards against over-reading descriptive numbers.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (world): Linking text mentions to WordNet/Wikidata entries could follow this pattern, with a cheap retriever proposing a short candidate list and an explainable chooser either picking one or recording "unlinked" rather than guessing.
- **I2** (world): Ontology nodes could be indexed with all their synonyms, glosses and known misspellings so that noisy mentions find the right concept.
- **I3** (world): Graded lexical properties such as concreteness or "is a measure" could be computed against small, documented anchor lists and stored as explainable features in the compiled knowledge.
- **I4** (learn): When building silver data, the LLM teacher tier could be allowed to return "no decision", and those cases could be excluded rather than counted as training labels.
- **I5** (learn): Each evaluation report could carry an explicit "does not show" note, much like a negative control, so that scores from a non-representative test set are not mistaken for general ability.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Netherlands Organisation for Applied Scientific Research](https://md.research.elamur.ai/free/en/labs/netherlands-organisation-for-applied-scientific-research/) — Research institute, Delft, Netherlands · 2 works

📚 Shelves (6): [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,772 works · [🤝 Sociology and Political Science](https://md.research.elamur.ai/free/en/shelves/social-sciences/sociology-and-political-science/) · 577 works · [🌿 Environmental Science](https://md.research.elamur.ai/free/en/shelves/environmental-science/) · 9 works · [📈 Economics, Econometrics and Finance](https://md.research.elamur.ai/free/en/shelves/economics-econometrics-and-finance/) · 17 works · [📈 Economics and Econometrics](https://md.research.elamur.ai/free/en/shelves/economics-econometrics-and-finance/economics-and-econometrics/) · 12 works · [🌿 Management, Monitoring, Policy and Law](https://md.research.elamur.ai/free/en/shelves/environmental-science/management-monitoring-policy-and-law/) · 2 works

## Cite

Cite as: Sadegh Shahmohammadi, Xavier Pinho, Mairi Bowdler, Suhendan Adiguzel-van Zoelen and Joost van Genabeek. 2026. Measuring Digital Labour Market Transitions with a Digital Semantic Score: An AI-Based Methodology Applied to the Dutch Labour Market. arXiv:2608.24222.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{shahmohammadi2026measuring,
  title = {{Measuring Digital Labour Market Transitions with a Digital Semantic Score: An AI-Based Methodology Applied to the Dutch Labour Market}},
  author = {Sadegh Shahmohammadi, Xavier Pinho, Mairi Bowdler, Suhendan Adiguzel-van Zoelen and Joost {van Genabeek}},
  year = {2026},
  doi = {10.48550/arxiv.2608.24222},
  eprint = {2608.24222},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2608.24222}
}
```

- [← previous Measuring and Mitigating Shortcut Reliance in Language Models with Probe-Based Representation Entanglement](https://md.research.elamur.ai/free/en/papers/singh-2026-measuring-mitigating-shortcut-reliance-language/)
- [next → Measuring Distribution Shift in User Prompts and Its Effects on LLM Performance](https://md.research.elamur.ai/free/en/papers/seegmiller-2026-measuring-distribution-shift-user-prompts/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/36/)

## References

14 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Alekseeva, L., Azar, J., Giné, M., Samila, S. & Taska, B. (2021), ‘The demand for AI skills in the labour market’. Labour Economics 71, 102002. [find](https://scholar.google.com/scholar?q=Alekseeva%2C+L%2E%2C+Azar%2C+J%2E%2C+Gin%C3%A9%2C+M%2E%2C+Samila%2C+S%2E+%26+Taska%2C+B%2E+%282021%29%2C+%E2%80%98The+demand+for+AI+skills+in+the+labour+market%E2%80%99%2E+Labour+Economics+71%2C+102002%2E)
2. Beblavý, M., Fabo, B., & Lenaerts, K. (2016). Demand for digital skills in the US labour market: the IT skills pyramid. [find](https://scholar.google.com/scholar?q=Beblav%C3%BD%2C+M%2E%2C+Fabo%2C+B%2E%2C+%26+Lenaerts%2C+K%2E+%282016%29%2E+Demand+for+digital+skills+in+the+US+labour+market%3A+the+IT+skills+pyramid%2E)
3. Engberg, E., Görg, H., Lodefalk, M., Jaced, F., Längkvist, M., Monteiro, N., Nordas, H., Pulito, G., Schroeder, S., Tang, A. (2024). AI Unboxed and Jobs: A Novel Measure and Firm-Level Evidence from Three Countries. IZA Discussion Paper DP No. 16717, Bonn, Germany. [↗](https://doi.org/10.2139/ssrn.4691172)
4. Escudero, V., Liepmann, H., & Podjanin, A. (2024). Using Online Vacancy and Job Applicants' Data to Study Skills Dynamics. [↗](https://doi.org/10.1108/s0147-91212024000052b023)
5. European Commission. (2026). [find](https://scholar.google.com/scholar?q=European+Commission%2E+%282026%29%2E)
6. Digital Decade 2026 country report: The Netherlands (Commission Staff Working Document SWD(2026) 155 final, Part 20/27). European Commission. [https://ec.europa.eu/newsroom/dae/redirection/document/128656](https://ec.europa.eu/newsroom/dae/redirection/document/128656) • • • • • Frey, C. B., & Osborne, M. A. (2013/2017). ‘The future of employment: How susceptible are jobs to computerisation?’ Technological forecasting and social change, 114, 254-280. [↗](https://ec.europa.eu/newsroom/dae/redirection/document/128656)
7. Green, A. (2024), “Artificial intelligence and the changing demand for skills in the labour market”, OECD Artificial Intelligence Papers, No. 14, OECD Publishing, Paris, https://doi.org/10.1787/88684e36-en. [↗](https://doi.org/10.1787/88684e36-en)
8. Jensen, M. F., & Schneider, C. (2026). Digital skill adoption and employment dynamics: Evidence from matched vacancy-employer–employee data. Research Policy, 55(5), 105476. • • LinkedIn. (n.d.). About LinkedIn. LinkedIn News. https://news.linkedin.com/about-us • • Lennon, C., Zilian, L. S., & Zilian, S. S. (2023). Digitalisation of occupations— Developing an indicator based on digital skill requirements. Plos one, 18(1), e0278281. [↗](https://doi.org/10.1016/j.respol.2026.105476)
9. Mason, C. M., Chen, H., Evans, D., & Walker, G. (2023). Illustrating the application of a skills taxonomy, machine learning and online data to inform career and training decisions. The International Journal of Information and Learning Technology, 40(4), 353-371. [↗](https://doi.org/10.1108/ijilt-05-2022-0106)
10. Napierała, J. (2024). Enhancing taxonomy-based extraction: leveraging information from online community platforms for digital skills demand identification in job ads. Statistical Journal of the IAOS, 40(3), 591-602. [↗](https://doi.org/10.3233/sji-230110)
11. OECD (2023), OECD Employment Outlook 2023: Artificial Intelligence and the Labour Market, OECD Publishing, Paris, https://doi.org/10.1787/08785bba-en. [↗](https://doi.org/10.1787/08785bba-en)
12. Revelio Labs. (2025). Revelio Labs Workforce Data [Proprietary dataset licensed to TNO]. Statistics Netherlands (CBS). (2025). Increasing use of AI by business. CBS. [https://www.cbs.nl/en-gb/news/2025/09/increasing-use-of-ai-by-business](https://www.cbs.nl/en-gb/news/2025/09/increasing-use-of-ai-by-business) [↗](https://www.cbs.nl/en-gb/news/2025/09/increasing-use-of-ai-by-business)
13. Stephany, F. (2021). One size does not fit all: Constructing complementary digital reskilling strategies using online labour market data. Big Data & Society, 8(1), 20539517211003120. [↗](https://doi.org/10.1177/20539517211003120)
14. Webb, M. (2020), ‘The impact of artificial intelligence on the labour market’, SSRN Electronic Journal. • • • • • • • [find](https://scholar.google.com/scholar?q=Webb%2C+M%2E+%282020%29%2C+%E2%80%98The+impact+of+artificial+intelligence+on+the+labour+market%E2%80%99%2C+SSRN+Electronic+Journal%2E+%E2%80%A2+%E2%80%A2+%E2%80%A2+%E2%80%A2+%E2%80%A2+%E2%80%A2+%E2%80%A2)

*Updated 1 November 2026*