Towards structural Web Services matching based on Kernel methods

Abstract

This paper describes a kernel methods based Web Services matching mechanism for Web Services discovery and integration. The matching mechanism tries to exploit the latent semantics by the structure of Web Services. In this paper, Web Services are schemed by WSDL (Web Services Description Language) as tree-structured XML documents, and their matching degree is calculated by our novel algorithm designed for loosely tree matching against the traditional methods. In order to achieve the task, we bring forward the concept of path subsequence to model WSDL documents in the vector space. Then, an advanced n-spectrum kernel function is defined, so that the similarity of two WSDL documents can be drawn by implementing the kernel function in the space. Using textual similarity and n-spectrum kernel values as features of low-level and mid-level, we build up a model to estimate the functional similarity between Web Services, whose parameters are learned by a ranking-SVM. Finally, a set of experiments were designed to verify the model, and the results showed that several metrics for the retrieval of Web Services have been improved by our approach.