Abstract

This project is developing an AI-powered job-matching system that uses modern machine learning techniques to improve the match between job seekers and open job postings. The system is based on large-scale labor market data containing high-dimensional skill information from job postings and individual profiles, as well as historical employment matches. Since real-world labor market data only captures successful transitions into employment, a so-called Positive Unlabeled Learning (PU Learning) approach is employed. The system combines skill profiles, job requirements, and other characteristics such as education or work experience to predict individual matching probabilities between individuals and job openings. The quality of a match, in terms of employment stability and expected wages, is also taken into account. Furthermore, regional labor market conditions and “labor market tightness” effects are considered to avoid competitive situations and overconcentration on individual job postings. This is intended to result in a recommendation system that is more economically efficient and realistic in terms of the labor market.