Web Service Discovery – A Reality Check Daniel Bachlechner Katharina Siorpaes Digital Enterprise Research Institute Digital Enterprise Research Institute Innsbruck, Austria Innsbruck, Austria email@example.com firstname.lastname@example.org Holger Lausen Dieter Fensel Digital Enterprise Research Institute Digital Enterprise Research Institute Innsbruck, Austria Innsbruck, Austria and Galway, Ireland email@example.com firstname.lastname@example.org ABSTRACT they will not scale without significant mechanization of service Web services are about the integration of applications via the discovery, service adaptation, negotiation, service composition, Web. Hereby, the programming effort should be minimized service invocation, and service monitoring; as well as data, through the reuse of standardized components and interfaces. protocol, and process mediation . Web services are service Semantic Web services try to provide the next step through endpoints in such architecture. If the SOA paradigm succeeds mechanizing important sub tasks within a service-oriented there will be soon several thousand services, which can be used architecture. Otherwise, significant manual programming effort for composing required applications. However, for this, these would remain as a bottleneck for this approach. One of the sub services must first be discovered. Within the semantic Web tasks in a service-oriented architecture is service discovery. While community, many of the publications on service discovery tend to a significant number of papers have already been published in this place more emphasis on certain aspects of reasoning rather than area, most of them are more concerned in providing yet another on focusing on current constraints and foreseeable evolvement of illustration for an arbitrary logical framework rather than service discovery (cf. , ). The survey summarized on the providing a contribution that meets current constraints in given poster takes the opposite approach. We enumerate existing practical settings. On the poster, we provide a comparison of approaches for public Web service discovery, compare them with existing approaches towards Web service discovery based on respect to specific criteria and identify their strength and empirical findings. This sets the basis for analyzing the strengths weaknesses. Based on the evaluation’s results we conclude and weaknesses of the existing approaches as well as the potential paths for semantics in Web service discovery as an prediction of future potential improvements in this area. We also extension of current approaches. identify a useful role for semantic techniques as long as it is in a proper setting. 2. SURVEY Based on previous work  we have identified several approaches Categories and Subject Descriptors for Web service discovery that are actually deployed and exceed D.2.11 [Software Engineering]: Software Architecture – data the scope of a dozen test services. We have examined the standard abstraction, domain-specific architectures, information hiding, UDDI registry approach, search via specialized portal sites and languages, patterns. customized searches using standard Internet search engines. General Term 2.1 Current Approaches Measurement, Experimentation The first of the three described approaches for current Web service discovery is based on UDDI. UDDI (Universal Keywords Description, Discovery, and Integration) is a standard for Semantic Web Services, Discovery, Service-oriented Architecture centralized repositories. The first UDDI Business Registry (UBR) nodes were run by IBM, Microsoft, SAP and NTT Com. 1. INTRODUCTION Examples of the second approach are specialized portals Service-oriented architectures (SOA) emphasize that it is the which gather services using focused crawlers as well as manual service that counts for the customer, not the specific software or registration. The list of Web service engines investigated within hardware component that is used to implement it. SOAs will the scope of the study includes XMethods, BindingPoint, likely become a leading software paradigm quickly. However, WebServiceX.NET, Web Service List, StrikeIron, Woogle, RemoteMethods, and eSynaps. This list of engines includes to our Permission to make digital or hard copies of all or part of this work for knowledge all relevant portals of the time of writing. Some personal or classroom use is granted without fee provided that copies are repositories known from previous work like SalCentral and Grand not made or distributed for profit or commercial advantage and that Central could not be accessed during the time of the survey and copies bear this notice and the full citation on the first page. To copy hence have not been evaluated. otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. The third approach uses standard Web search engines which Demos and Posters of the 3rd European Semantic Web Conference are able to restrict the search to WSDL files. We analyzed the (ESWC 2006), Budva, Montenegro, 11th - 14th June, 2006. search engines Google and Baidu with respect to their ability to facilitate and enable Web service discovery. Google and Baidu 4. CONCLUSIONS have different means to restrict search queries to specific types of Based on our findings, searching with Google has the best documents, and given the huge size of the underlying document coverage, although the precision is limited since there is no single index, both likely to be big players in the long-run. way to restrict a search to only retrieve active and working services supposed to test examples. Most of the public UDDI 2.2 Criteria registries have been discontinued in early 2006, however, due to The criteria used for the evaluation can be classified into two the limited quality of the contained data, for searching public groups, the first of which represents basics for Web service services they have never been a good source. All existing discovery and deals with core criteria like the ways of how a specialized Web search engines provide less coverage than search can be conducted, number of available services, status Google. However, the standard model of Google is not well suited information, and supported interfaces. The second criteria group for Web service discovery. Neither the identification of potential consists of service rating, test and demo features (like WSDL services through pure key word extraction nor the relevance analyzer), and service costs which allows service discovery to be ranking based on HTML characteristics such as hyperlinks and more precise and less time-consuming. title tags provides much of a use in a Web service scenario. The usage of standardized vocabulary such as UNIFACT or eClass to 3. SURVEY RESULTS classify Web services could significantly improve the correctness The Web service resources presented on the poster follow many and completeness and do not provide much of a burden to Web different approaches of service discovery with varying success. service providers. If needed, this task can be mostly automated by The findings of the evaluation are represented in tabular form. approaches such as GoldenBullet . Furthermore the page The table describes the current approaches in Web service ranking mechanism of Google that uses the link structure and discovery in terms of the introduced evaluation criteria. The three special properties of HTML documents are not applicable to groups of resources as well as the two groups of criteria are WSDL files. Therefore, different post processing and filtering clearly separated. We used the developed evaluation scheme to mechanisms of the output of Google are needed. This is a task conduct an in-depth comparison of the discovery approaches. where richer semantic annotations can play a role. The majority of the approaches relies on keyword search as Simple application of IR technologies and later the use of well as category browsing whereas XMethods only shows ontologies to describe standard vocabulary are the most promising services in a simple list format. The UBR also allow searching for approaches for the near future. Rich formal frameworks are service providers and tModels. The Web service search engine required as well, but should be considered more in the scope of Woogle additionally offers template search on operations. semi closed environments (e.g. extra nets), where full automation Obviously the state of the art of search functionality is rather is possible. For the near future the role of central portal providers limited and hampers usability. Semantic Web services could will most likely become more important in the domain of public enable a more comprehensive search as well as automation of services: Take Amazon as an example. The effort in maintaining tasks. Especially in the UBR, the location of Web services is and developing this central repository is high but it is profitable difficult as it does not provide an efficient interface for querying too. StrikeIron for instance follows a similar business model. services. The name of a Web service, a Web service provider or a However in the long run the business model might be invalidated tModel must be known to get further details. The UBR keyword by the advancement of technology: If semantic Web service search only takes names into account and ignores service technologies advance it is likely that such intermediates (between descriptions. Considering service descriptions could be promising service provider and consumer) will loose its current importance. in theory, but unfortunately most of the descriptions available are deficient and of low quality. Due to the limited extent of human 5. REFERENCES readable descriptions in the UBR, discovery is a cumbersome and  Akkiraju, R., Goodwin, R., Doshi, P. and Roeder, S.: A time-consuming effort. Most numbers provided by Web service method for semantically enhancing the service discovery search engines, concerning the number of registered Web capabilities of UDDI. In S. Kambhampati and C.A. services, are vague and imprecise. However, it is obvious that Knoblock, eds., Proceedings of the IJCAI-03 Workshop on Google provides a significantly higher number of WSDL files. Information Integration on the Web, pages 87-92, 2003. The number of available services in terms of a specific  Benatallah, B., Hacid, M-S., Rey, C. and Toumani, F.: discovery provider is an important indicator for the Request rewriting-based Web service discovery. In The comprehensiveness of a Web service discovery engine. However, Semantic Web - ISWC 2003, pages 242-257, October 2003. at large, service functionality and quality are of course far more important than quantity. Some Web service resources provide  Ding, Y., Korotkiy, M., Omelayenko, B., Kartseva, V., functionality to determine whether a service is active or not. The Zykov, V., Klein, M., Schulten, E. and Fensel, D.: UBR does not provide any status information at all while GoldenBullet in a Nutshell. In Proceedings of the 15th StrikeIron and Woogle display the status of a listed Web service International FLAIRS Conference, Pensacola, USA, 2002. (active or inactive). BindingPoint allows for excluding inactive  Fan, J. and Kambhampati S. A Snapshot of Public Web Web services from its listings. Another helpful piece of Services, SIGMOD Record, 34(1), pages 24-32, March 2005. information provided by BindingPoint in this context is the average response times of specific services. All evaluated  Fensel, D. and Bussler, C.: The Web Service Modeling resources for locating Web services have Web interfaces. Selected Framework WSMF. In Electronic Commerce Research and ones also provide SOAP and UDDI Private Registry interfaces as Applications, 1(2), pages 113–137, 2002. well as RSS feeds, WS-Inspection.