We mined 10 years worth of data on 100 million interactions between job counsellors and job seekers to develop a recommender system - allowing job counsellors to make smarter, more personalised suggestions on which interventions unemployed individuals should take.
The goal of this project was to help the Institute of Employment and Vocational Training in Portugal (IEFP) to reduce the time job seekers spend unemployed in continental Portugal. The unemployment rate (as the number of people as a percentage of the labour force actively looking for work) in Portugal is 6.5% as of August 2019, this compares to 3.7% in the UK, and 3.2% in Germany (OECD, 2019). Unemployment has serious consequences for individuals’ psychological and physical health (Wanberg, 2012).
Images: Rosa Lavelle-Hill presenting the prototypes at Data Fest 2019
The European Union has been calling state members to develop tools that could help individuals find job placements quicker (Scoppetta, & Buckenleib, 2018). One of the barriers that may hinder job candidates’ ability to find a job is a mismatch between the individual’s skills and the skills required by the available jobs. Currently, job counsellors provide service users with recommendations for an intervention. This is typically done by selecting an intervention from a large list of potential interventions, in a limited amount of time. Furthermore, the job counsellor’s knowledge about which interventions are most effective for a given individual in a given labour context is limited by their previous experiences. Therefore, the recommendations counsellors give are not always optimal.
We mined 10 years worth of data on 100 million interactions between job counsellors and job seekers. We developed a prototype of a recommender system that would augment the capabilities of IEFP job counsellors to propose interventions to the unemployed. The system provides personalised and relevant interventions, taking into account contextual information about the individuals’ skills, aspirations, restrictions, and their socio-economic context. These recommendations support the collaborative process of building personal development plans for the service users, and are intended to help in finding suitable employment.
Figure 1. Model Selection in 2 stages: first of the supervised model, and second of the recommendations produced using ground truth examples in 10 years of historical data.
When scaled up, the individual-level benefits could have a cumulative net effect of reducing nation-wide unemployment rates in mainland Portugal. The personalised recommendations also benefit job counsellors, providing them with a more diverse but also more targeted set of interventions to be considered. This reduces the cognitive pressure during the development of the personal plan, and introduces tractability of which options have been considered during the process. Other benefits of the recommendation system include the provision of high level insights about the effectiveness of particular interventions which can enable smarter data-driven decisions on funding to be made.
References
OECD (2019), Unemployment rate (indicator). doi: 10.1787/997c8750-en (Accessed on 13 August 2019)
Scoppetta, A. & Buckenleib, A. (2018). Tackling long-term unemployment through risk profiling and outreach. a discussion paper from the employment thematic network. European Commission - ESF Transnational Cooperation. Technical Dossier no. 6. Luxembourg: Publications Office of the European Union, 2018.
Wanberg, C. R. (2012). The individual experience of unemployment. Annual review of psychology, 63, 369-396.