A lightweight approach for classifying job titles in an economic context

Abstract
The classification of a company’s employees job titles into different departments can be used to obtain the best approximation of a company’s internal structure from an external observer’s point of view. This paper describes several approaches to create a compressed, lightweight data set which will be used to train a Support Vector Machine for classifying job titles into 15 different departments. The scope is set to primarily handle job titles, which can be found in an industrial setting. In this paper, employees of eleven companies belonging to the industry sector of mechanical engineering had been screened for job titles to work with. The goal is to have a data set of unambiguous, unique terms, where each term can exclusively be assigned to one of the 15 given departments. While creating the training data set, an approach for identifying the hierarchical position of a job title will also be shown, which helps to depict the exact position of an employee in an organizational structure. The evaluation of the developed model clarifies the great potential of using mentioned approaches for job titles which are as unambiguous as possible. The highest accuracy, which could be achieved during the evaluation process of the model, peaked at nearly 90% and has even more potential, when exclusively using unambiguous job titles for testing.