Artificial intelligence for ESCO maintenance process

Updating ESCO requires making informed decisions based on an evidence-based approach that integrates data from several sources such as expert input, national occupations and skills classifications, job seekers profiles, courses and online job advertisements. 

In recent years, the European Commission initiated an investigation into the potential applications of artificial intelligence (AI) in supporting the work of experts in maintaining, improving and simplifying the use of ESCO. Over time, these techniques were integrated into the mainstream ESCO maintenance process, resulting in the release of ESCO v1.1 in February 2022. The methodology was subsequently further developed for the release of minor version ESCO v1.1.1 and major version 1.2. 

On a broader level, the use of artificial intelligence within the context of the ESCO project can be understood as the range of data science-related tasks that comprise the following work strands: 

• Data collection and processing: Data from sources such as national occupations and skills classifications, Europass profiles, courses, domain experts’ input and online job advertisements are combined and extracted in order to provide an up to date set of labour market data to inform the update of ESCO. 
• Core methodology development: Essential algorithms and artificial intelligence models are developed for deriving insights from raw data and for shaping these data such that the ESCO maintenance process can directly benefit from it. The core methodology is used to identify potentially new skill/occupation concepts, provide quantitative evidence for new/existing concepts (e.g. usage of a concept across various data sources), extract and validate links between skills and occupations and detect quality issues. 
• Quality performance management: A significant component of developing algorithms, models and artificial intelligence tools is having the infrastructure and processes in place to measure their quality and to be able to measure progress over time. This component is an integrated part of the overall core methodology development approach. 
• Applied data science: Once specific methodological building blocks have been developed, they are integrated and executed in specific processes.

The development of base models is a continuous effort and the Commission is constantly running experiments to improve them and measure progress over time. The underlying reasoning for this is that improvements in these building blocks positively impact many applications. Additionally, the work strand on data collection and processing provides repeated data updates (e.g. expert validation) which allows to improve the quality of base models over the course of time. Examples of base model development are approaches to suggest relevant ESCO occupations for concepts from multilingual national occupation classifications, and a methodology to suggest ESCO skills for learning outcomes in course descriptions as part of the Learning Outcome Linking Pilot Project. The different base models contain multilingual skill extraction, multilingual skill mapping, multilingual skill classifier and finally clustering similar phrases. 

Moreover, task-specific models are developed to achieve more specific goals, i.e. components having a more narrow range of applications but still being central to the work in ESCO. Two examples of task-specific models are NACE classifier and ISCED-F classifier.

Additional documentation

• Technical Report: Leveraging Artificial Intelligence to maintain the ESCO Occupations Pillar
• Technical Report: Leveraging Artificial Intelligence to update the ESCO Occupations Pillar 
• Data Science and ESCO | ESCO blog post