A Review of AI-Based Resume Screening and Job Matching Systems

Abstract

Recruiters today are facing a challenge. They get many applications that its impossible to review them all. Because of this resume screening has become the automated part of the hiring process. This paper looks at how intelligence and machine learning can be used to screen resumes and match candidates to jobs. We examine methods such as keyword extraction, TF-IDF similarity, document embeddings, Siamese convolutional networks and transformers. We focus on four studies: a systematic review of AI in hiring a TF-IDF semantic matching approach, a deep Siamese network for resume-job matching and a survey of machine learning techniques in resume screening. As the way we represent data gets more advanced—from simple word counts to semantic models—the accuracy of matching improves. Most of the models we studied only do one thing. They don't test their results against hiring decisions. They rely on labels. They don't explain why a candidate gets a certain score. The results on reducing bias are mixed. Issues like transparency and privacy remain unsolved. We believe that a better solution is needed—a system that is integrated, explainable and aware of bias and that includes people in the loop. To address this we introduce the HireSense AI framework. We also suggest directions for research.

Keywords