PaperlayerView exact revision on arXiv

Source-linked AI summary

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

Sadegh Shahmohammadi, Xavier Pinho, Mairi Bowdler, Suhendan Adiguzel-van Zoelen, Joost van Genabeek

arXiv:2608.24222v1cs.CL

TL;DR

Rapidly evolving technologies make it difficult to identify and classify emerging digital skills with existing taxonomies. This study introduces a Digital Semantic Score using large-scale profile data, finding that digitalisation is uneven across occupations and mobility pathways.

  • Problem

    Rapidly evolving technologies challenge the identification and classification of emerging digital skills using existing taxonomies and classification systems.

  • Method

    The study combines large-scale online profile data with a Digital Semantic Score measuring digitalisation in occupational language and worker skill profiles.

  • Results

    Digitalisation is occupationally uneven, with robust gains concentrated in knowledge-intensive and managerial work while selected technical and operational occupations show diffusion.

  • Takeaways & Limitations

    The Digital Semantic Score enables measurement of digitalisation through occupational language and worker skill profiles rather than predefined groups or keyword-based indicators.

  • Takeaways & Limitations

    Skills are person-level profile attributes rather than job-specific observations, so they do not directly measure skills used, acquired, or required in particular jobs.

Abstract

from arXiv · show

The digital transformation of the Dutch labour market is reshaping occupational language, career pathways, and job-related skills. Addressing these changes requires granular labour market intelligence. This paper develops an AI-based methodology to analyse digitalisation using data covering millions of Dutch job profiles. The methodology combines embedding-based similarity search and large language model classification to map unstructured job information to harmonised ESCO occupations. We also introduce a Digital Semantic Score that measures how strongly job titles and skills are associated with digital concepts relative to a non-digital reference. Using embeddings and cosine similarity to transparent digital and non-digital anchor groups, this indicator moves beyond keyword-based approaches by capturing broader digital meanings in occupational language and worker skill profiles. It enables analysis across occupations, career transitions, emerging job-title vocabulary, and skill digitality. The findings reveal that digitalisation is unevenly distributed across the labour market. Digital job-title language is most prominent among managerial, professional and ICT-related occupations, but is increasingly visible in hybrid business, marketing and automation-related roles. Career-transition analyses show that movement toward digital work is pathway-dependent, while skill analyses highlight the multidimensional nature of digital capability, encompassing technical, hybrid and business-systems skills. By combining profile data, AI-supported occupational classification and semantic scoring, this study advances AI-driven labour market analytics and provides a scalable framework for monitoring digital labour market change. The methodology helps identify emerging skill needs, support reskilling strategies, and inform policies addressing skills mismatches and labour shortages in the Netherlands.

Introduction

Digitalisation, automation and AI are rapidly transforming the Dutch labour market, changing occupational tasks, skill demand and the emergence of labour and skills shortages. The study responds with granular online-profile data, AI-supported occupational classification and a Digital Semantic Score for analysing digital labour-market change.

  • Introduction: In 2024, 22.7% of Dutch businesses with more than 10 employees integrated one or more types of AI technology.These developments are changing occupational tasks, skill demand and contributing to emerging labour and skills shortages.
  • Introduction: Existing research often relies on broad occupational classifications, vacancy data or aggregate statistics that miss changing work language, new job titles and skill transferability.The introduction identifies these as important measurement challenges in studying digitalisation.
  • Introduction: Granular online labour-market data, worker profiles and natural language processing can reveal changing skill requirements, occupational dynamics and labour-market mismatches beyond conventional sources.These approaches support analysis at greater detail and pace than many conventional labour-market data sources allow.
  • Introduction: Rapidly evolving technologies such as AI create continuing challenges for identifying and classifying emerging digital skills within existing taxonomies and classification systems.Big-data labour-market analytics and AI-supported text classification offer opportunities to address these limitations.
  • Introduction: The study classifies large-scale Dutch online profile data into ESCO-based occupations and introduces a Digital Semantic Score measuring the semantic association of job titles and skills with digital concepts.Its contributions include AI-supported occupational classification, measurement of digital content over time, and application to variation across occupational groups and emerging job-title vocabulary.

Data and Methods · Dataset

The study analyses Dutch job-history data from a large-scale online profile dataset, combining occupational, skills, employment-history and educational information. The dataset supports broad labour-market analysis but cannot link self-reported skills to specific jobs or track skill development longitudinally.

  • Dataset: The underlying Revelio Labs dataset comprises over one billion job profiles, 20 million companies and more than 3,000 reported skills.It also includes job histories and educational information.
  • Dataset: Profile fields are partly extracted directly from LinkedIn and partly predicted by Revelio Labs.Job titles and skills are directly extracted, whereas gender and remote suitability are predicted.
  • Dataset: The analysis focuses specifically on Netherlands job-history data.The Netherlands is described as a high-LinkedIn-adoption market with a rich digital footprint.
  • Dataset: The Netherlands ranks consistently among the highest markets globally for LinkedIn adoption.This supports the availability of extensive online labour-market profile data for the country.
  • Dataset: The Netherlands ranks top in the European Union’s Digital Economy and Society Index, with over 83% of the population having basic to advanced digital skills.The paper links this context to an unusually rich workforce digital footprint.
  • Dataset: Each unique user has a list of self-reported skills, enabling analysis of individual skill profiles.However, those skills cannot be linked to specific jobs within employment histories.
  • Dataset: The data cannot identify which skills corresponded to particular occupations or how skills evolved across job transitions over time.This prevents individual longitudinal analyses of skill development and constitutes a dataset limitation.

ESCO integration

The methodology reclassifies free-text jobs into Europe-relevant ESCO occupations through embedding retrieval and language-model selection, then harmonizes assignments to four-digit ISCO codes. It also measures digitality as relative semantic proximity to transparent digital anchor groups versus a non-digital baseline.

  • ESCO classification: A two-stage AI pipeline embeds job descriptions and ESCO entries, retrieves ten candidates with FAISS, and uses a language model to select the most appropriate occupation.Embeddings use OpenAI’s text-embedding-3-large model with 3,072-dimensional representations; low-confidence cases receive no occupational code.
  • ISCO harmonization: Assigned ESCO occupations are harmonized to four-digit ISCO-08 codes, with invalid four-digit codes excluded from ISCO-based analyses.The ISCO major group is identified by the first digit of the retained four-digit code.
  • Occupational transitions: Job-to-job transitions are formed from chronologically consecutive records within individuals and retained only when both source and destination occupations have valid codes.Each consecutive pair contributes one observed occupational transition.
  • Digital Semantic Score: The Digital Semantic Score compares a title or skill’s cosine similarity to three digital anchor groups against similarity to a non-digital baseline, rather than using keywords or supervised classification.The digital groups are core_digital, hybrid_digital, and business_systems_automation; the score is a relative semantic index, not a percentage or probability.
  • Digital Semantic Score: Subtracting the non-digital baseline prevents broadly work-related or professional language from being mistaken for digital content and can produce low or negative scores.A text may be closest to a digital dimension yet remain closer to non-digital anchors overall.
  • Digital Semantic Score: The four concise anchor sets are transparent and reproducible reference points, not a complete taxonomy of digital and non-digital work.The present results use exactly the listed anchor groups, while future robustness checks could vary or expand them.

Anchor Group Definitions and Rationale

The methodology uses manually defined, transparent semantic anchor groups to establish interpretable embedding-space directions for three forms of digital work and a non-digital baseline. Broad English anchor phrases represent semantic fields, while separate digital-domain similarities and baseline contrast support differentiated digital semantic scoring.

  • Anchor groups: Manual anchor groups define interpretable embedding-space directions rather than supervised labels or outcome-estimated references.The groups represent core technical digital work, hybrid digital work, business systems and automation, and a non-digital baseline.
  • Anchor groups: Broad English anchor phrases represent work domains rather than individual tools or narrow occupations, reducing dependence on single keywords.Job-title text was translated to English, while available skill strings were normalized into Englishlike lower-case terms.
  • Core digital: The core_digital group captures technical digital production and infrastructure, including software, data, AI, cloud, cybersecurity, databases, web systems, IT, and business intelligence.A title or skill close to this centroid is interpreted as semantically close to core digital work.
  • Hybrid digital: The hybrid_digital group captures digitally mediated business, marketing, product, analytics, customer-facing, CRM, UX, and e-commerce work outside necessarily core technical IT roles.Its phrases include digital transformation, digital marketing, online platforms, product ownership, agile management, customer analytics, CRM systems, user experience design, and e-commerce operations.
  • Business systems and automation: The business_systems_automation group captures enterprise software, administrative digitization, process and workflow automation, RPA, enterprise platforms, low-code automation, and data-driven operations.A title or skill close to this centroid is interpreted as close to business process digitalization or automation.
  • Baseline and scoring rationale: The non_digital_baseline group provides a contrast direction for generally less digital manual, service, craft, logistics, and administrative work without implying those jobs never use digital tools.Subtracting baseline similarity helps isolate whether a title is more digital than general non-digital work, while the three digital groups remain separate and the strongest digital similarity is used.

Job-title digitalization over time

Job-title digitalization is summarized by assigning each job record its normalized title’s Digital Semantic Score and averaging these scores by year and ISCO major group. These averages describe observed title semantics without external employment weighting.

  • Measurement: Each job record inherits the Digital Semantic Score of its normalized job title.
  • Interpretation: The yearly major-group averages represent observed title semantics and are not adjusted using external employment weights.

Four-digit occupational digital drift

Four-digit ISCO occupational digital drift compares rowweighted mean title scores between 2000–2010 and 2020–2024. Rankings exclude or flag occupations with low support, improving descriptive stability without constituting a formal significance test.

  • Comparison periods: Four-digit ISCO occupations are compared across the 2000–2010 baseline period and the 2020–2024 recent period.The comparison summarizes longer-run occupational change using digital title scores.
  • Drift measure: Occupational digital drift is defined from changes in rowweighted mean title scores across the two periods.The method specifies a digital drift measure for each occupation.
  • Weighting: The rowweighted mean title score uses the number of observed job records for each occupation and period.The observation count is denoted by n_gq for occupation g during period q.
  • Support and limitations: Occupations with low support in either period are flagged or excluded from reportable rankings.Support filtering improves descriptive stability but is not a formal statistical significance test.

Transition-level semantic change

Transition-level semantic change is measured for same-person moves between source and destination jobs, then summarized across occupational groups. Visualizations include positive, near-zero, and negative changes to represent the full range of transition patterns.

  • Transition-level measurement: Same-person transitions measure the change in title digitalization from source job u to destination job v.The measure is defined at the individual transition level.
  • Occupational-group aggregation: Transition summaries aggregate semantic changes by source and destination occupational groups.For each transition group A→B, the summary uses the mean semantic shift.
  • Visualization: Balanced transition visualizations report positive, near-zero, and negative transition groups rather than only the largest increases.This presentation captures changes across the full direction of transition-level semantic shifts.

Skill-level analysis

Skill-level analysis applies the anchor-based Digital Semantic Score to normalized person-level skill strings, but interprets results as profile-level associations rather than year-specific occupational skill requirements.

  • Method: Individual skills are parsed into distinct normalized strings and assigned the same anchor-based Digital Semantic Score as job titles.The score is applied at the person level to each skill in the individual’s skill list.
  • Method: Person-level skill digitalization is averaged across each individual’s parsed skill set.The average is defined over the set of parsed skills for person p.
  • Interpretation: Skill analyses measure profile-level associations, not direct year-specific occupational skill requirements.Skill lists are not linked to each specific job.

Results & Discussion

The results show that labour-market digitalisation is uneven across occupations and career pathways, while increasingly shaping both established occupational language and newly emerging job titles. The findings also indicate that digital skill signals are broad rather than limited to rare or highly specialised terms.

  • Occupation-level digitalisation: Managers and professionals have the highest average digital scores, with the strongest increases concentrated in managerial, professional, ICT-related and technical-professional occupations.Digital terminology has become more embedded in higher-skill and knowledge-intensive occupational language, including management, consulting, administration and specialised professional roles.
  • Occupation-level digitalisation: Technicians, service and sales workers, clerical support workers, craft workers and machine operators generally remain near or below zero, although some non-professional groups show meaningful increases.These scores indicate semantic proximity to the non-digital baseline rather than absence of digitalisation, and moderate-support changes remain suggestive.
  • Career transitions: Transitions from elementary occupations to managers show an average shift of approximately 0.117, with 89% positive transitions, while transitions to professionals show approximately 0.111 and 83% positive transitions.The strongest positive mobility shifts are concentrated in pathways into managerial and professional occupations.
  • Emerging job-title vocabulary: Newly appearing job titles become increasingly digital over time, indicating that digitalisation shapes both changes within established occupations and the creation of new occupational labels.Each normalized title is counted once in its first observed year, separating vocabulary emergence from job-row volume; the 2024 title-count decline may reflect partial-year coverage and data sparsity.
  • Digital skills: High-prevalence skills are ranked by digital semantic score only when listed by at least 500 people, highlighting common digital capability signals rather than rare specialised terms.The threshold focuses interpretation on skills that are both semantically digital and sufficiently common in observed profiles.

Conclusion

The study concludes that digitalisation is uneven across occupations, career pathways and skill profiles, extending beyond traditional ICT roles into hybrid work. Its Digital Semantic Score and AI-supported methodology provide a scalable basis for monitoring change and informing reskilling policy, while limitations require cautious interpretation.

  • Conclusion: Digitalisation differs across occupations, career pathways and skill profiles, extending beyond traditional ICT occupations into hybrid professional, managerial and operational roles.The findings also identify emerging occupational vocabulary and mobility pathways connecting occupations with different digital profiles.
  • Conclusion: The Digital Semantic Score measures digitalisation through occupational language and worker skill profiles without relying on predefined occupational groups or keyword-based indicators.This provides a distinct analytical contribution for examining the spread of digitalisation.
  • Limitations: LinkedIn-based data are not representative of the Dutch workforce and overrepresent highly educated, white-collar, professional, managerial and knowledge-intensive occupations.Manual, routine, lower-skilled and less digitally oriented occupations are underrepresented in the dataset.
  • Limitations: Profile-level skills cannot be linked to specific jobs or career transitions, so skill analyses reflect individuals’ overall digital orientation rather than changing job-specific requirements.The data cannot establish when skills were acquired, which skills were used in particular jobs, or how portfolios changed across transitions.
  • Limitations: Self-reported skills may be incomplete, outdated, inconsistently labelled, or overstated or understated, despite normalization and semantic embeddings reducing some measurement error.These limitations affect interpretation of the skill information.
  • Implications: The methodology provides a scalable framework for monitoring digital labour market change, identifying emerging skill needs, and supporting evidence-based reskilling and labour market policies in the Netherlands.Future research could further examine the evolution of skills at the job level.

Supplementary Files

Figure S1 examines how observed 4-digit ISCO career transitions connect occupations with different average digital skill profiles. It shows that digital labour-market change follows specific, multidirectional pathways rather than a uniform mobility gradient.

  • Method: Figure S1 compares origin and destination occupations using average digital semantic scores of worker-reported skills, rather than job-title scores.The analysis aggregates person-level skill embeddings to occupations and summarizes each transition by the difference between destination and origin skill digitality.
  • Method: Displayed pathways include the largest positive shifts, near-zero changes, and lowest or negative shifts, with at least 100 transitions and 50 workers per pair.Both origin and destination occupations had to include workers with skill scores for the comparison.
  • Positive pathways: Positive shifts lead toward advertising and marketing professionals, sales and marketing managers, and systems analysts, including hybrid roles combining domain expertise with digital tools and analytics.The destinations extend beyond narrowly defined core IT occupations to roles involving platforms, CRM systems, automation, and data-driven decision-making.
  • Positive pathways: Some positive pathways originate in service, retail, clerical, and administrative occupations such as waiters, shop keepers, accounting associate professionals, and administrative secretaries.This indicates that movement into more digitally oriented environments need not proceed only between technical occupations.
  • Mobility patterns: Near-zero and negative shifts show that mobility is not unidirectional: some moves connect similarly digital occupations, while others lead from digitally skilled to less digitally characterized environments.Near-zero examples include transitions between software developers and advertising and marketing professionals; negative cases may reflect career, management, domain, data, or temporary-work factors.
  • Limitation: The figure does not demonstrate that workers gained or lost digital skills during transitions; it describes the average skill digitality associated with origin and destination occupations.Skills are person-level attributes rather than skills linked to each individual job.
Loading 2608.24222v1…