A Two Phase Approach for Efficient Clustering of Web Services

Sorting out a desired web service is a demanding concern in service oriented computing as the default keyword search options provided by UDDI registries are not so promising. This paper deals with a novel approach of employing an unsupervised neural network based clustering algorithm namely ART (Adaptive Resonance Theory) for service clustering. The input to the algorithm includes both functional characteristics which are quantified using the basic user requirements in phase 1 and non functional characteristics which are derived by means of swarm based techniques through appropriate mapping of metadata to swarm factors and thereafter updating the input in phase 2. Taking the advantages in being an unsupervised clustering algorithm, ART in a more potential way groups services eliminating a number of irrelevant services returned over a normal search and facilitates to rearrange the registry. Clustering depends on a threshold value namely vigilance parameter which is set between 0 and 1. Flocking of birds is the swarm behaviour considered.