Recommendation for Newborn Services by Divide-and-Conquer

Abstract

Service recommendation plays a critical role in fostering the growth of service ecosystems. However, existing methods are mainly in favor of a small number of popular services while newly emerged ones (i.e., newborn services) are largely ignored, which hurts the systems in two aspects. First, the potential of many services, especially the newborn ones, is wasted. Second, service ecosystems highly depending on a few kernel services are not diversified nor robust. To address this issue, we propose to proactively recommend collaborative services for newborn ones. The aim is to illuminate how to use the newborn services and fertilize their proper usages. While this is a cold start problem, frequent collaboration among newborn or dissimilar services makes it more difficult. In this situation, a Divide-and-Conquer approach is adopted utilizing category tags and collaboration records (DCCC). For each newborn service, the approach first produces one ranked list of old services and one list of newborn services, separately. DCCC then merges the two lists into one for recommendation. Experiments over a real-world dataset from ProgrammableWeb demonstrate that the proposed approach achieves significant improvement in recommendation accuracy compared with baseline methods.