Data Science Ph.D. Qualifier Presentation | Marcela Vasconcellos | Monday, July 10th, 2023 @ 3:00PM
3:00 p.m. to 4:00 p.m.
United States
DATA SCIENCE
Ph.D. Qualifier Presentation
Marcela Vasconcellos
Monday, July 10th, 2023 | 3:00PM
Location: Unity Hall 471
Zoom: https://wpi.zoom.us/j/95878842441
Committee:
Dr. Andrew C. Trapp, PhD Advisor
Dr. Oren Mangoubi, Co-Advisor
Dr. Randy Paffenroth, Co-Advisor
Title:
Fair Many-to-Many Matching of Refugees to Jobs with Ties, Incomplete Lists, and Elastic Lower and Upper Quotas via Integer Optimization
Abstract:
The vast majority of the over 103 million displaced persons around the world are in vulnerable situations due to a lack of long-term solutions for their inclusion in host communities. I consider the challenge of securing stable employment for at-risk refugees, presently handled in a cumbersome, manual manner. I formulate the problem of matching refugees to job opportunities as a many-to-many, cardinal preference-based matching problem with ties, incomplete lists, and elastic lower and upper quotas. I present an integer optimization formulation for this problem that delivers generated solutions as recommendations within a job-matching platform. My formulation considers both refugee preferences and needs, as well as job requirements. While integer optimization formulations for the one-to-one and many-to-one preference-based matching problems have seen successful applications, this work is the first to formulate one in the many-to-many context. I investigate the effects of using upper and lower bounds on the number of matches per candidate as a means to improve the welfare of more vulnerable candidates. I also investigate the effects of including fairness requirements and extend the state of the art of the integer optimization matching literature by developing novel sets that in the many-to-many context eliminate justified envy, a key component of stability. The proposed experiments are tested on synthetic, representative profiles of refugees and job opportunities. Results show improvement in refugee outcomes in the presence of lower bounds and successful elimination of justified envy with the proposed constraints with only minor deterioration of the objective function value.