Ranking résumés automatically using only résumés: A method free of job offers

Abstract With the success of the electronic recruitment, now it is easier to find a job offer and apply for it. However, due to this same success, nowadays, human resource managers tend to receive high volumes of applications for each job offer. These applications turn into large quantities of documents, known as resumes or curricula vitae, that need to be processed quickly and correctly. To reduce the time necessary to process the resumes, human resource managers have been working with the scientific community to create systems that automate their ranking. Until today, most of these systems are based on the comparison of job offers and resumes. Nevertheless, this comparison is impossible to do in data sets where job offers are no longer available, as it happens in this work. We present two methods to rank resumes that do not use job offers or any semantic resource, unlike existing state-of-the-art systems. The methods are based on what we call Inter-Resume Proximity, which is the lexical similarity between only resumes sent by candidates in response to the same job offer. Besides, we propose the use of Relevance Feedback, at general and lexical levels to improve the ranking of resumes. Relevance Feedback is applied using techniques based on similarity coefficients and vocabulary scoring. All the methods have been tested on a large corpus of 171 real selection processes, which correspond to more than 14,000 resumes. The developed methods can rank correctly, in average, 93% of the resumes sent to each job posting. The outcomes presented here show that it is not necessary to use job offers or semantic resources to provide high quality results. Furthermore, we observed that resumes have particular characteristics that as ensemble, work as a facial composite and provide more information about the job posting than the job offer. This certainly will change how systems analyze and rank resumes.