Two-Step System for Identifying Job Titles in Online Job Postings

Abstract

Data science approaches are very useful for mining massive databases for insights. There has been a lot of recent focus on employment market analysis via the categorization of internet job adverts.

In order to successfully determine the job title from an advertising, many multi-label classification methods have been created, such as clustering and self-supervised learning. These methods, however, are only applicable to US-centric databases like O*NET and need labelled datasets containing hundreds of thousands of cases. To tackle the issue of tiny datasets, we provide a two-stage job title identification mechanism in this study. The job advertising is first classified according to their appropriate sector (e.g., Information Technology, Agriculture) using Bidirectional Encoder Representations from Transformers (BERT). After that, we locate the job title that most closely matches the projected industry using a combination of similarity metrics and unsupervised machine learning methods. New tests confirm that the suggested approach is superior than or on par with the state-of-the-art approaches. By using the suggested approach with data from the Moroccan labour market, we were able to spot new, in-demand jobs in the country.