Today we released version 1.0 of the Bright Score, a machine learning algorithm that we are very proud of. The Bright Score is the result of the largest resume to job description scientifically controlled clinical trial in history. We will post a short academic paper on the score soon, and we expect to publish our conclusions in a peer-reviewed academic journal in the fall. The goal of our scientists developing the Bright Score was to emulate optimal human behavior of a talent recruiter during the resume evaluation process.
To date, matching systems of resumes to job descriptions relied on keyword pattern matching, a notoriously unreliable measure of the qualifications of an applicant against a job description. Other systems have relied almost exclusively on synonym sets to identify similar terms that may not be known outside an area of expertise (e.g., java and J2EE).
While the Bright Score factors in keywords and synonyms, alone those features are not sufficient to discriminate between candidates at scale. The Bright Score algorithm is informed by sophisticated features, including historical relationships of resume features (e.g., employer, skill, school, job titles) across our resume database. More specifically, features of the Bright Score include, but are not limited to, managerial relationships and industry taxonomies; inverse document frequencies based upon resume and job description corpuses; gaps in employment and job-hopping; if an applicant is overqualified; previous versus current salary expectations; career trajectory; skills along career trajectory; company prestige; if an applicant previously worked for a competitor; required and desired skills; certifications; school rank; education timeline; several different semantic relationships between the resume and job description; resume and job description spectral density; social fingerprint; company connections; social network; personality traits; cognitive profile; unstructured internet data; unique analysis of data from the Bureau of Labor and Statistics and many other available sources; SIC codes; SEO; etcetera.
The data set for the Bright Score was over 8.6 million job seekers that visited our lab (Bright.com), 2.1 million job descriptions, 2.8 million resumes, and over 100 talent recruiter human evaluators.
Please continue to check back here for more about the science behind the Bright Score.
David Hardtke, Ph.D.; Chief Scientist