Online Ranking of Physicians with Perishable Resources

Abstract

For the online healthcare platform, it is essential to recommend a list of physicians to arrived patients dynamically, in order to satisfy patients’ varied demands, improve the overall matching degree between patients and physicians as well as the resource allocation efficiency, since each physician only has limited registration capacity. In this paper we construct a general model to describe this dynamic ranking problem with patients’ rank-based choice behavior and limited resources. Under the framework of Markov Decision Process(MDP), the dynamic problem can be decomposed into a series of static problems as long as the marginal value of each physician’s registration resource can be approximated. To deal with the problem that at the cold-start stage the platform does not have available data to capture patients rank-based choice behavior specifically, we design a model-free LP-based method which can efficiently approximate the marginal value without requiring patients’ choice models as the input. We conduct a case study based on the data collected from the online healthcare platform to show how to apply this general framework in a real application and verify the capability of our methods.