Lun LI, Ph.D. Candidate in Data Science

Bio

Lun Li, a Ph.D. student in data science at Stevens Institute of Technology, specializes in large language models, predictive modeling and dynamic graph modeling.

Skillsets

She is skilled in Python, pySprak, R, SQL, MongoDB, Alteryx, Google BigQuery, Redshift and Unix.

Dissertation Summary

Workforce Analytics in the Era of Big Data

This dissertation comprises three essays to develop innovative and systematic methods for workforce analysis. In particular, by utilizing large-scale online resume data and advanced deep learning technologies, we examine one of the most prevalent topics in workforce analysis – attrition and retention – at both the organizational and individual levels.

Organizational Level: The first essay introduces a novel algorithm, TalentRank, to assess and track employers’ attractiveness. This algorithm offers a dynamic ranking system based on the changes in talent inflow and outflow, and it provides detailed rankings across various job titles. As a by-product, TalentGrouper is devised to address the significant variability encountered in online resume databases, making it a valuable tool for other workforce analysis.

Individual Level: The main thrust of the second and third essays is to explain individuals’ job-hopping behavior via a theory-driven deep-learning paradigm. Under the guidance of category theory, we refine the definition of typicality to account for the temporal axis, which captures the evolving expertise of an individual and thus their ability to acquire a job that they are most experienced in. Although dynamic typicality plays a significant role in explaining occupational mobility, it does not reconcile the fact that individuals do span the boundaries for “untypical” opportunities. To this end, we mainly examine the following aspects to further explain the residuals,

– Inter-Occupational Similarity: This metric is introduced to facilitate occupation mobility among those that require similar skillset, or skills that are easily transferable.

– Individual Preference: We hypothesize that individuals with high career diversification are consistently inclined to seek new job opportunities, regardless of job type. This behavior is quantified using a diversification indicator, which tracks their propensity for pursuing varied career options.

– Social Network Influence: The career choice of each individual is also influenced by their peers. We introduce a dynamic graph model that connects the focal employee to potential occupations through reference employees with similar career trajectories.

In summary, this work holds significant implications for employers to optimize their human capital strategies and provide a data-driven solution to retain talents more effectively. We conduct extensive experiments and compare our approach to competitive baselines to demonstrate how it can improve the results of key workforce analysis applications.

Academic Advisor

Rong (Emily) Liu