Behavioral Similarity of Semantic Web Services?

Abstract. Service matchmaking is an integral link of service discovery, composition, invocation and other similar task under Service-Oriented Architecture (SOA). Most current approaches measure the degree of match of two service based merely on their I/O pairs which could leads to false result. This paper presents an approach for matchmaking in Se-mantic Web Services (SWS) that considers each service as a sub-graph of semantic network, which is formed by inputs, outputs, pre-conditions and post-condition, with contribution of syntactical information such as keywords from the service description. Thus the similarity between ser-vices is defined as the similarity between two sub-graphs. The aim of this approach is to reveal the internal work flow and intention of services, i.e. behaviors, thus it agrees with human intuition to a larger extent than previous approaches. 1 Introducction The original intention of adding semantic annotations to web services is to improve the automation of services discovery, selection, invocation and inter-operation by letting service descriptions to be machine-processable [7]. One in-tegral part of such automation is matchmaking among services. Various approaches have been proposed in previous studies. Without con-cerns about semantics of its components, one primitive method to calculate the similarity of services is based on the syntactical information- e.g. keywords, tag-clouds and textual descriptions. For services with semantic information, inputs/outputs (I/O) matching is a common method for measuring the similarity. Inputs and outputs of a semantic service are instances of ontological concepts, the similarity of two services is determined by the minimal distance in the taxonomy tree between corresponding concepts of I/O pair, the result is a degree of semantic simialrity, such as exact, plug-in and subsumes [6]. Some studies, such as [5], aimed to achieve higher robustness and precision by combining both. More recently, various graph based approaches have been proposed. In [4], a service was considered as a composition of processes and thus could be rep-resented as a finite-state machines (FSMs), the similarity between services wa