Client Clustering for Hiring Modeling in Work Marketplaces

Abstract

We study the problem of grouping clients of an online work marketplace into clusters, such that in each cluster clients are similar with respect to their hiring criteria. Such a separation allows the marketplace to “learn” more accurately the hiring criteria in each cluster and recommend the right contractor to each client, for a successful collaboration. We provide a Maximum Likelihood definition of the “optimal” client clustering, based on a logit model. To find the optimal clustering, we propose a scalable Expectation-Maximization algorithm. We study the baseline behavior of the algorithm using synthetic data and we verify that our approach yields significant gains compared to the baseline approach of “learning” the same hiring criteria for all clients, using a real dataset of 865,000 hiring decisions provided by oDesk. In addition, we analyze the clustering results on the oDesk dataset and we find interesting differences between the hiring criteria of the different groups of clients.