Service Clustering with Graph Embedding of Heterogeneous Networks

The API economy, characterized by the growing number of Web services, necessitates effective service discovery and management methods. A commonly used method is service clustering, categorizing Web services based on functionalities described in WSDL files. However, this approach is highly dependent on the description and susceptible to variations in naming conventions and languages employed by service providers. Therefore, we propose a novel method of service clustering that leverages service dependencies in composite services. We represent atomic services and their co-invocation relationships as a graph and apply graph embedding techniques for service clustering. Our method extends the existing node2vec technique by considering composite service contexts to embed features of atomic services more accurately. Experimental application of our method on a real-world Web service dataset demonstrated a 10% increase in clustering accuracy, as measured by the purity metric, compared to the traditional description-based method.