21th AIAI 2025, 26 - 29 June 2025, Limassol, Cyprus

Improving the Diversity and Fairness in Job Recommendations using the Stable Matching Algorithm

Unecha Jatin, Moh Melody, Moh Teng-Sheng

Abstract:

  In the current competitive job market, recommendation systems are essential to connect job seekers with suitable work opportunities. For job recommendation systems, ensuring fairness, reducing application congestion, and enhancing diversity remain ongoing challenges. Traditional models such as collaborative filtering and content-based methods often focus primarily on job seeker preferences, resulting in over-recommendation of popular roles while underexposure of others. In this work, we propose an adapted application of the Gale-Shapley stable matching algorithm to address these issues. The new approach introduces configurable quotas for both applicants and jobs, enabling scalable many-to-many matchings while considering mutual preferences. We empirically evaluate our method using real-world job application data and compare it with Matrix Factorization and ReCon, a congestion reducing model recently proposed by Mashayekhi, et al. Results show that the proposed approach achieves significant improvements in diversity (coverage), fairness (Gini index), and congestion metrics, while maintaining comparable accuracy. To the best of our knowledge, this is the first job recommendation system that adopts the stable matching approach. The promising results reported in this work underscore the great potential of a stable matching-based approach to enhance the fairness and effectiveness of recommendation systems, particularly for two-sided matching applications, in which the preferences for both sides should be considered for a successful, stable matching.  

*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.