All Times EDT
Keywords: topic models, job matching, variational methods
The related tasks of finding suitable candidates for a job and suitable positions for a job seeker are important for both organizations and individuals. Current solutions to these problems are predominantly manual, based on keyword searches, which is both time-consuming and ignores a large part of the rich information available . In this work, the correlated topic model (CTM) is used to identify properties of job description and job seeker resume pairs that were successfully matched. Given the fitted CTM, a comparison of topic distribution between resumes and job descriptions using Kullback-Leibler divergence allows one to identify resumes that match a given job, or jobs that match a given resume, Our approach facilitates the 'many to many' matchings that are a feature of such data. We apply our approach to a data set consisting of 583 unique job descriptions, 53,688 unique resumes and over 75,000 matched connections, and obtain on average a 20-fold improvement over baseline performance when the top 100 ranked matches are considered. The scaleability of the method is also explored.