Abstract: Embodiments of the present invention are directed to grouping clients in a work marketplace into clusters such that, in each cluster, clients are similar with respect to their hiring criteria. In some embodiments, the clusters are generated based on a clustering algorithm that can be applied effectively on large datasets. This separation allows the work marketplace to discover differences in client hiring criteria, to learn more accurately the hiring criteria in each cluster and to recommend the right contractor to each client for a successful collaboration, thereby improving matching between clients and contractors in the work marketplace. For each contractor who have submitted an application to a project posted by a client, a contractor/opening pair score is determined based on information of the cluster that is associated with the client. The contractor/opening pair score is used to ascertain whether the contractor would be a candidate recommended to the client.