Deep selection: inferring employee traits from résumé style using neural networks

Abstract We explored the predictability of individual factors related to job performance (e.g., cognitive ability and Big Five personality traits) based on applicant’s résumé formatting that was inferred by a computer. Humans typically emphasize the content of the résumé, such as education, work experience, and skills listed, when subjectively evaluating the applicant. Because job applicants may falsify or exaggerate their résumé information to match the prospective employer’s desires, using the content to draw inferences about the applicant has limitations. To address this limitation, we trained an artificial neural network to learn the common visual features within a résumé. We used these visual features as predictors in a machine learning model to predict an applicant’s work-related traits on a sample of 387 résumés. We found that the visual features of a résumé could predict both people’s personality traits and their cognitive ability beyond chance levels. We compared those predictions with the ratings made by two human raters using a linear regression model that predicted the true trait scores with the human rater scores and the neural network model’s predictions. Results show that neural networks significantly improved the explanatory ability of the model compared to only using human ratings. The relationship between résumé style and people’s traits could not be explained by simple measures of visual appearance such as the word count or darkness of content. Using this scalable method, practitioners can extract additional information from their existing selection process to better identify qualified job candidates within large pools of applicants.