Fairness-aware Matching for Dynamic Predictive Behavior Matching

Web-based chat services are widely used by millions of customers today. However, few of them satisfy both service providers and recipients (customers).For instance, customers are continually complaining about long wait times and inadequate service. The technical challenge is to keep both the customers and the agent happy. To solve these difficulties, we plan to model predictive matching as a dynamic Learning-to-Rank problem. In order to maximize the success rate of matches between agents and consumers, we offer a dynamic fairness-aware learning-to-rank approach that takes into consideration agents’ fair working restrictions. Regarding assessment, we plan to develop a new, unbiased A/B testing procedure that takes into account the unique limits of our contact center operations. Our system is aimed to achieve Pareto optimal in terms of both precision and fairness compared to other systems. The project will provide novel solution for deploying fairness customer-agent behavior matching technique into real-world applications.

Faculty Supervisor:

Xue (Steve) Liu

Student:

Partner:

Bell Canada (QC)

Discipline:

Computer science

Sector:

Information and cultural industries

University:

McGill University

Program: