Intelligent System for Workforce Classification

We examine several machine learning algorithms for assigning job-categories to a large employee population based on their online directory profiles and reporting structures. This application is an important step in workforce management, and also represents a challenging classification problem which typically contains large-scale unbalanced data, i.e., the distribution of job categories is highly skewed. It also requires combining local information, e.g., the textual description in an employee’s webpage, with global structure, e.g., the reporting chain structure, in order to predict job categories. We compare the performance of graphical models to several classification and relational learning algorithms with well-tuned features and show satisfactory accuracies on two datasets of more than 20,000 IBM employees. We also address some challenges that the graphical models face when applied to large-scale data.