Comparing Transformer Models in Their Ability to Screen the Best Applicants for a Job Search

Most job searches today involve an online component, where companies traditionally rely on keyword searches to match applicants to job descriptions. With the rise of transformer models, these advanced tools are being applied in new areas, including human resources. In this article, we compare traditional keyword searches on CVs to three transformer models—BERT, RoBERTa, and GPT-3—using a set of job descriptions. As expected, the transformer models, which can assess context, outperformed traditional keyword searches. However, the performance differences among the three transformer models were minimal. We explore the reasons behind this result and discuss potential strategies to enhance the performance of these models in future job screening processes.