Big Data-driven Talent Analytics: A Deep Learning Approach

Talent analytics refers to using big data, statistical modeling, and machine learning techniques to address workforce-related decision-making problems. As one challenge in talent analytics, employee turnover is one of the most critical challenges for companies, especially those ones in Information Technology (IT) industry that confront high turnover rates. To address this challenge, we develop a novel deep survival analysis model to predict individual employees’ turnovers across different companies. The developed approach makes important methodological contribution to the literature in threefold. First, we propose a formalization of discrete-time survival analysis for the turnover prediction problem. Rooted in theoretical foundations, we operationalize four groups of time-dependent driving factors within the survival analysis framework. Second, we develop a novel deep survival analysis model that effectively models the constructed time-dependent factors, temporal dependency between hazard rates over time intervals, and the competitor influence between firms that affects employees’ turnover decisions. Third, based on the model design, we propose a learning problem that maximizes the likelihood of discrete-time hazard rates and introduce a mini-batch stochastic gradient descent algorithm to estimate the model parameters. With extensive experimental evaluations with a real-world dataset, we demonstrate the superior performance of our method over several benchmark methods, and further demonstrate its practical business values with case studies. Other than employee turnover, I briefly introduce a new study on talent skill gap (funded through an NSF CAREER Award) and our proposed novel design science artifact, namely CampusLens, with a variety of career, education and job data collected from multiple sources.