Big Data for Labour Market Intelligence in Tunisia — Machine-Classified Job Vacancies for VET Planning

Big Data for Labour Market Intelligence is a system run by the European Training Foundation (ETF), an agency of the European Union headquartered in Turin, and implemented for and with Tunisia's Ministère de la Formation Professionnelle et de l'Emploi. It scrapes online job vacancies and uses natural-language processing and machine learning to classify each advertisement into ISCO-08 occupations at four-digit level, ESCO 1.1.1 skills, NACE sectors and ISCED-2011 education levels. The classified data feed public dashboards intended for skills anticipation and for designing vocational education and training provision. Technical partners are Lightcast and CRISP at the Università degli Studi di Milano-Bicocca. The methodology deliberately mirrors Cedefop's Skills-OVATE so that results are comparable across countries. Tunisia was one of the first two full-scale pilots, its dashboard has been live since 2020, and collection continued monthly as of 2025.

The ETF's Tunisia analytical working paper of 16 November 2020 reports 37,535 online job vacancies collected for Tunisia between 1 April and 30 September 2020. Supply is highly concentrated: TANITJOBS provided more than 19,000 advertisements, about 52 per cent of the total, JORA around 5,500 or 15 per cent, KEEJOB around 4,800 or 13 per cent, and EMPLOINAT around 2,200 or 6 per cent. Twenty-six per cent of advertisements do not state a contract type. Permanent contracts accounted for 42.4 per cent of advertisements at the "Professionals" skill level against 31.5 per cent at lower skill levels. Across the whole programme, an ETF presentation of June 2025 reports 23 million vacancies collected and 10 million retained after deduplication, dashboards for six countries — Egypt, Morocco, Tunisia, Kenya, Ukraine and Georgia — in three languages, five capacity-building programmes between 2019 and 2024, and 45 training videos.

The methodological documentation is the strongest feature. There is a published analytical report with a stated method, a live public dashboard, and an underlying peer-reviewed literature base on machine classification of job advertisements. The ETF publishes its own limitations unusually candidly: online-vacancy sampling bias, with some sectors over-represented and others entirely absent; a heavy requirement for specialist expertise; and unstructured data needing standardisation. The Tunisia paper adds that web advertisements are not representative of self-employment, a serious caveat in a labour market with large informal and self-employed segments.

What is missing is any evidence of use or effect. Nothing published shows how many Tunisian VET programmes, qualifications or guidance services were actually changed because of the dashboards, and the number of revised programmes is not published. Operational ownership inside Tunisia could not be confirmed from ETF documents, leaving institutionalisation and long-term sustainability open questions and creating a real risk that the system remains donor-run rather than embedded. The 2020 Tunisia paper is itself labelled a workshop draft whose contents do not necessarily reflect the views of the ETF. This is best read as a well-documented, honestly caveated data infrastructure whose downstream educational effect remains unmeasured.

Read the full analysis: https://openspace.etf.europa.eu/content/big-data-labour-market-intelligence