Towards Matching Maximization in Job Recommendation System with Reinforcement Learning

Abstract

This study explores a recommendation system using reinforcement learning for a job-matching system. Job recommendation systems have different properties than typical e-commerce recommendation systems that maximize user click-through rates or purchases since they aim to maximize the number of matches between users and jobs under the constraint that each job has its own position quota. Conventional job recommendation systems have proposed such methods as distributing jobs to which users apply or using mathematical optimization algorithms. Still, these methods require specific assumptions and strong conditions and do not directly maximize the number of matches. In this study, we propose a method to directly maximize the number of matches by using reinforcement learning that incorporates the number of matches as a reward. Simulation experiments on synthetic data demonstrate that the proposed method improves the number of matches by at least 131.2% compared to existing methods, confirming its superior performance.