Matching Résumés to Job Descriptions with Stacked Models.

We describe a method for matching resumes to job descriptions provided by employers, and evaluate it on real data from a Canadian company specialized in e-recruitment. We model the task as a classifying each resume as suitable or not for a follow up interview. We evaluate the methods on two datasets with approximately 1,500 real job descriptions and approximately 70,000 resumes, from two important industry sectors, considering several models individually and also stacked. Our stacked model shows high accuracy (often above 0.8) and consistently outperforms standard methods, including neural networks.