NOTATIONS
4.5 Conclusion
output parameter patterns used for the first set of experiments. The experiment were repeated by publishing 1000 web services in service registry. A comparison of time taken for output parameter pattern based search prior to clustering and after clustering is depicted in figure 4.8. From the graphs obtained we can infer that on integrating our clustering approach intoextended service registry, there is a substantial improvement in the performance of parameter based service search.
(a) registry with 500 services (b) registry with 1000 services
Figure 4.8: Output Parameter based Search Using Clusters
Thus, in general measuring similarity between two parameters has an importance for preprocessing of service registry to find clusters of services. In order to widen the scope of search for a web service with a given set of input/output parameters, it is useful to cluster services having similar input/output parameters. Hence, we propose an ap- proach for clustering service input/output parameters on their similarity, using WordNet as the underlying ontology.
AlgorithmSCSP in section 4.2.3 generates parameter clusters of input/output pa- rameters of the web services in a registry. The cluster selection in the algorithm is gov- erned by the values ofandminSP, as discussed in section 4.2. We have simulated the algorithms on QWS Dataset(Al-Masri and Mahmoud (2008)). Experimental evidence shows that the clusters generated by our algorithm has a better Precision and Recall values when compared with the clusters generated by K-means algorithm(MacQueen et al.(1967)).
We utilize these parameter clusters for clustering services with similar input/output parameters to enable efficient parameter based web service search in service registries.
Further, we propose an approach to accelerate service search of non-semantic web ser- vices by clustering services on their output parameter similarity. Our approach for clustering web services based onfrequent output parameter patterns looks promising since it provides a natural way of reducing the candidate services when we are looking for a web service with desiredoutput parameter pattern.
Algorithm F OP C in section 4.3.2 generates a covering cluster that covers all the web services in the registry. The cluster selection in the algorithm is governed bypref- erence factor, as discussed in section 4.3.1. We have simulated the algorithms on QWS- WSDL dataset(Al-Masri and Mahmoud (2008)). The experimental results demonstrate the performance of our clustering approach on varying user queries.
Algorithm4 for service search discussed in section4.3.3 lists a set of web services that matches the queried output parameters. From the many services that are returned we need to choose the best match for a given user query. Hence, we propose a service selection approach that ranks the set of matching services based on their QoS values in the next chapter. Further a service composition method is proposed that can be used when the service search fails to find a web service matching the user requirements.
CHAPTER 5
Parameter based service selection and composition
Abstract
One critical challenge in web service search and composition is the selection of web services, to be executed or to be composed, from the pool of matching services. Most of the current service selection proposals (Cai and Xu (2014); El Hadadet al.(2010); Liet al.(2014b); Mobedpour and Ding (2013);
Yageret al.(2011)) apply a weighted sum model (WSM) as an evaluation method for selection of ser- vices with the same functionality. We propose abi-level service selection approach that selects the most appropriate web services from the pool of matching services that considers both the functional and non-functional requirements for service selection. The functional requirements are provided by a user as a set of input parameters provided for and output parameters desired from the web service. The user also provides a set of desired QoS values and the order of their preference for selection.
By service composition, we mean making of a new service (that does not exist on its own) from ex- isting services. Most of the existing algorithms for service composition consider services based on exact matches of input/output parameters for composition. However, for service composition the construction of such a chain of services fails at a point when the output parameters of a preceding service (wsOP) does not match exactly with the input parameters of a succeeding service(wsIS). We propose1to alleviate this problem by extending the classical definition of service composition to give rise to three types of service composition:exact composition,super compositionandcollaborative composition. Adopting three types of service composition for a desired service output parameter set, the possibility of having different kinds of compositions is demonstrated in form of acomposition search tree. Further, we propose2the utility ofcomposition search treefor finding compositions of interest likeleanest compositionand theshortest depth compositions.
1Lakshmi.H.N, Hrushikesha Mohanty;RDBMS for service repository and composition, Proceedings of 4th International Conference on Advanced Computing, ICoAC 2012, published in IEEE Xplore.
2Lakshmi.H.N, Hrushikesha Mohanty,Utility of composition search tree for searching Optimal ser- vice compositions, Published in Elsevier Science, proceedings of Eighth International Conference on Data Mining and Warehousing ISBN: 9789351072515, 2014 Pp 36- 42.