Skills Matching 2.0: Fair decision-making in the jobs market by using AI in recruitment processes

Published on: 11 March 2021

The Dutch labour market is struggling with friction in demand and supply of labour. Employers cannot find the employees they need, and employees experience they cannot use large parts of their knowledge and skills in their current job or are not be enabled to work more hours. Labour market surveys show that mismatches are only increasing, e.g. due to digitalization and automation, and with it the inequality of opportunities on the labour market (CBS, 2019; UWV, 2019). To remain relevant and employable, workers must continuously re-evaluate and update their skillsets; companies face a pressing need for innovative talent sourcing, matching and development strategies; and educators need to update the focus of their courses and offerings.

To understand and meet emerging skills demands and to empower individuals to learn, unlearn and relearn skills, a transition from a diploma based to a skills based learning and working system is needed. A prerequisite for a skills based learning and working environment is a common currency; a skills language that can recognize, certify, reward and enhance skills.

Matching job seekers and employers based on skills has the following challenges:

  • The skills ontology must be dynamic due to changes in skills needs;
  • The bias in skills descriptions by gender, age or ethnicity is socially unacceptable,a matching function of skills supply to skills demand must go beyond the current simple point scales to enable more effective matching;
  • Automated results of skills matching and bias mitigation functions must be explainable to users and these functions must interact with (the knowledge of) labour market experts for optimal results.

In the use-case Skills Matching we are investigation how the use of Artificial Intelligence (AI) can offer solutions to these points.

Use of artificial intelligence

Various AI innovations are being develloped within the project, such as hybrid AI, responsible AI and a skills ontology:

  • Hybrid AI: combining AI techniques (such as natural language processing to extract skills from job vacancy texts) with knowledge models (using what is known as a ‘skills ontology’ in which people’s expertise is stored). This is a prerequisite for the dynamic approach mentioned above.
  • Responsible AI: using AI to determine where bias and discrimination occur in job vacancy texts and actively giving suggestions to assist in monitoring discrimination in the labour market. AI is also used for reducing bias when matching jobseekers to employers on the basis of skills. This helps ensure that both the AI and the recruitment and selection process are fair.
  • Skills ontology: together with the Dutch Employee Insurance Agency (UWV) and others, work is being done on the AI innovations needed for an improved, semi-automated and dynamic skills ontology that factors in the latest status and trends in the jobs market.

What challenge does it solve?

The project aims to improve the matching of jobseekers to vacancies based on their skills, with as little bias as possible and with matches that make sense and can be explained. We are doing that by making use of the three AI innovations mentioned above. This should ultimately lead to more satisfied employees and employers and will bring down the numbers of jobseekers.

What will the use case teach us?

The results envisaged include:

  • A methodology that will make it possible to use hybrid AI systems on a skills ontology and that will represent the dynamic aspects of the labour market well.
  • A tool for matching employers to jobseekers, with matches that can be explained and with a focus on preventing employment discrimination.

First results

To get more insight in the potential of common language in a skills-based labor market, the presentation on the ECP yearly festival can be found here.

A paper describes the prototype developed in 2020 of a methodology that can improve the matching of vacancies and jobseekers based on skills and with as little bias as possible and where the matches are explainable. The developed demonstrator contains functions such as classification, skills extraction, bias detector and the algorithm to calculate the overlap of competences between a vacancy and the skills ontology. If you are interested in trying the demonstrator, please contact Merle Beaujon of TNO.

Next steps

In 2022 we will continue to work on the dynamic skills ontology, bias in vacancy texts and skills matching algorithms and skills matching.

With respect to making the skills ontology dynamic by learning from existing knowledge bases, in 2021 we focused on occupation knowledge bases and on reusing existing crosswalks for the learning. In 2022 we will apply NLP technologies, in order to replace the manual work that is currently common practice. Moreover, we focus on different knowledge to learn about, namely skills and qualification knowledge.

With respect to making the skills ontology dynamic by learning from up-to-date labour market data, in 2021 we focused on finding similarities between skills in the skills ontology and vacancy texts. In 2022 we strive for identifying new concepts for and integrating them in the skills ontology.

With respect to bias, we focus on the first step of bias mitigation, which is detection. Based on the results of the analyses and exploration done in 2021, in 2022, we only work on gender bias. We aim at better detection of gender bias both in vacancy texts and in skills matching algorithms.

With respect to matching, in 2021 we focused on finding matches between skills of job applicants and skills in vacancies and identifying the skills gaps. In 2022, we develop transition paths for workers and job seekers from low- to high-potential jobs based on skills.

Parties involved

This project is part of TNO’s Appl.AI programme and it is partly financed from the start-up fund that NL AIC received from the government for research and development of AI applications. The project, which is led by TNO, is a cooperative effort by several parties including the Employee Insurance Agency (UWV), Foundation for Cooperation on Vocational Education, Training and the Labour Market (SBB), Statistics Agency Netherlands (CBS), The Netherlands Institute for Human Rights (CRM) and Centerdata.

Pending articles

There are some articles pending publication:

  • S.Vethman, A. Adhikari, M. H. T. de Boer, J. A. G. M. van Genabeek, C. J. Veenman “Context-Aware Discrimination Detection in Job Vacancies using Computational Language Models”, being submitted to the ACM conference of Fairness, Accountability and Transparency 2022 (link researchgate)
  • S. Vethman, A.F. van Luenen, J.A.G.M. van Genabeek, C.J. Veenman, “Gender imbalance on the job market reflected bias in skills descriptions in major skills ontologies”, being submitted to “Science Advances” (AAAS). (link to AAAS website)
  • T. Zoomer, J.A.G.M van Genabeek and L. Oosterheert, “Leren van skills en het verkrijgen van ander werk: Een verkenning van de mogelijkheden voor leren en employability bij beroepsgroepen”, submitted to Tijdschrift voor Arbeidsvraagstukken (Dutch Scientific Journal on labour issues), 2022 theme issue on the dynamics on the Dutch job market (DNA VI). (link to website)

Share via: