A Deep Learning BERT-Based Approach to Person-Job Fit in Talent Recruitment

Abstract

Although the widespread use of the Internet provides job recruiters with a larger pool to select the most qualified candidates, the tedious process of going over hundreds of resumes makes a fair and objective decision making more difficult. This paper proposes an end-to-end BERT-based framework to decrease the workload and expedite the shortlisting process of job applicants. Utilizing historical-records data of thousands failed and successful job applications, our model simulates the recruiters’ decision-making process by the state-of-the-art BERT algorithm. The results show that BERT outperforms a variety of models by a high margin.