related work to LOG4SWS.KOM and COV4SWS.KOM. Hence, these approaches have been chosen as benchmarks in the evaluation (cp. Section 3.4).
Most approaches make use of logic-based similarity measurement between concepts in an ontology.
Apart from the path length, alternative means to derive the semantic relatedness of concepts are not applied. Even though a lot of different approaches have been proposed, the missing integration of all ser-vice abstraction levels is noticeable. Most matchmakers presented are restricted to the serser-vice signature, in some cases the whole service profile (including preconditions and effects) is regarded. This might be traced back to the fact that still extensive, heterogeneous ontologies are missing which are needed to semantically describe service components [77, 103]. Therefore, we integrate fallback strategies in LOG4SWS.KOM and COV4SWS.KOM, which are usually missing (with the work of Syeda-Mahmood et al.
and Plebani and Pernici being notable exceptions).
The integration of different service abstraction levels further raises the question on how to weight the different levels (cp. Section 3.2.1). Here, the automatic adaptation as conducted in COV4SWS.KOM seems to be a good choice. In comparison, the adaptive matchmakers presented by Kiefer et al. and Klusch et al. are adaptable regarding the combination of different similarity measures. As a last aspect, it should be noted that runtime performance is usually not regarded explicitly. In LOG4SWS.KOM and COV4SWS.KOM we make use of caches and ATSM in order to boost the runtime performance.
H2: The computation of similarity metrics on all service abstraction levels instead of only considering the service signature leads to an improvement of matchmaking results. However, the weighting is im-portant: Regarding the applied test data set, levels which feature only syntactic descriptions should be integrated but weighted to a lesser extent than those levels which feature semantic descriptions of service components. However, the actual ideal weighting depends on the similarity metrics applied – in case of COV4SWS.KOM, manual weighting performs better than the automatically derived ones, except for the similarity metric proposed by Resnik.
H3: Even though the evaluation results for COV4SWS.KOM are usually lower than those of LOG4SWS.KOM, similarity metrics from the field of relatedness measurement of semantic con-cepts in ontologies are still performing well, especially if compared to other matchmakers (cp. Ta-ble 3.6). Most importantly, the automatic weighting of similarity values from different service abstraction levels in combination with the metric proposed by Resnik leads to the best evaluation result regarding AP for any SAWSDL matchmaker, so far.
Of course, there are possible extensions of the work conducted: To start with, it would be interesting to in-tegrate preconditions and effects into the matchmaking process of LOG4SWS.KOM and COV4SWS.KOM.
While the number of services observed in our evaluation and the competitive runtime performance of our matchmakers does not make it necessary to apply more sophisticated ways to downsize the search space, this could be necessary if following the vision of an “Internet of Services” where billions of ser-vices are available [72, 197]. Here, clustering or a top-down approach, which eliminates serser-vices not meeting certain requirements on interface and operation levels, could be integrated. Furthermore, it might be suitable to make use of alternative similarity measures or heuristics in order to detect similar-ities between service components. The same applies to matching algorithm – instead of the Hungarian algorithm applied in this thesis, another method might be more efficient if this is necessary. There might be further ways to enhance the retrieval performance of the matchmakers at hand – hence, we label the matchmakers presented in this thesis with LOG4SWS.KOM Version 1.0 respectively COV4SWS.KOM.
As a last aspect, most matchmakers neglect non-functional aspects like QoS which is also done in LOG4SWS.KOM and COV4SWS.KOM. However, an integration of the work of Berbner and Eckert would be straightforward and could easily be evaluated using an appropriate test collection [23, 74]. This could be done incorporating a semantic-based description of QoS characteristics as well as countermeasures if QoS requirements of a requester cannot be met [217, 223].
4 Query Formalisms for Semantic Web Services
In this chapter, the formulation of queries for service discovery is examined. Usually, approaches to syntactic service discovery rely on keyword search or category browsing, while queries aiming at SWS are mostly encoded using the same formalism as applied for the description of services, i.e., follow a
“query by example” approach. This makes it necessary to define a preferably complete description of the service capabilities desired, i.e., a service profile in OWL-S or a service interface in SAWSDL. As a result, query formats are mostly incompatible regarding different Web service standards. Further deficiencies of current query formats are the missing integration of syntactic and semantic search and the absent consideration of ranges of values.
The following hypotheses are applied in order to determine the outcome of the work conducted in this chapter. These hypotheses will be discussed in Section 4.6.
H4: Current query formalisms for SWS do not provide the necessary means to define requirements in a user-friendly, flexible, and precise way. Hence, it is necessary to deploy alternative means to formulate service requests. Already existing query languages could be a foundation for this, even if they have not been initially developed for application in SWS discovery.
H5: A unified query formalism for SWS should be independent from any particular registry or Web service specification. Thus, it is necessary to define a (unified) query language and corresponding service data and query models, which are per se independent from but nevertheless applicable to different standards.
This chapter is structured as follows: First, shortcomings of current approaches to query formulation for (semantic) Web services are identified (Section 4.1). Based thereupon, requirements towards service query languages will be specified (Section 4.2). Following this, two distinct approaches to query lan-guages for SWS will be described, which aim at different goals: The approach described in Section 4.3 has been the first approach designed and implemented in the context of this thesis. It aims at providing a lightweight integration of SPARQL-based queries for SAWSDL-based Web service descriptions into UDDI.
In contrast, the second solution (presented in Section 4.4) provides a more extensive approach to en-hance a service registry by alternative query languages. Therefore, an abstract query model is proposed.
Based thereupon, SWS2QL, an extension of SPARQL, aiming at SWS retrieval, is conceptualized. As proof of concept implementation, an ebXML-based service registry is enhanced. Before the chapter closes with a summary, Section 4.5 gives an overview on related work in this field of research.