Semantic matchmaking { i.e., the task of nding matching (Web) services based on semantic information { has been a prominent eld of research lately, and a wide range of supporting tools both for research and practice have been published. However, no suitable solution for the visualization of matchmaking results exists so far. In this paper, we present the Matchmaking Visualizer, an application for the visual representation and analysis of semantic matchmaking results. It allows for the comparing of matchmaking approaches for semantic Web services in a ne-grained manner and thus complements existing evaluation suites that are based on rather coarse-grained information retrieval metrics.
In the envisioned Internet of Services (IoS), organizations will be able to
realize business processes using not only internal (Web) services, but also external
(Web) services from public marketplaces. One prerequisite to this vision is that
functionally equivalent services can be e ectively and e ciently identi ed in
the potentially huge collection of available service o ers. This service discovery
process based on functional requirements is referred to as matchmaking. Recent
approaches to matchmaking frequently utilize semantic information in addition
to the purely syntactical service descriptions [1{3], based on the notion of
Semantic Web services (SWS) [
In current research, the evaluation of matchmaking approaches is mostly based on standard Information Retrieval (IR) metrics such as precision and recall. These metrics are, for instance, employed in the popular Semantic Matchmaking Evaluation Environment (SME2) tool1. IR metrics, however, exhibit a rather coarse-grained representation of matchmaking performance, because the service rankings are merged into a selected number of (more or less representative) gures. This type of evaluation provides little insight into the actual matching process. For instance, the similarities that have been assigned to certain pairs of components in a service, such as operations, and the resulting assignments can not be assessed in detail. However, this type of information can be bene cial
in order to comprehend a matchmaking process in detail. It can also help to analyze required modi cations to a (Web) service for actual consumption within, e.g., a complex business work ow.
In this paper, we present the Matchmaking Visualizer (MV) as a complement to existing matchmaking evaluation tools. It facilitates the visual representation and analysis of individual service o ers and requests and the respective matching results. This permits a much more ne-grained evaluation and assessment of matchmaking engines. 2
2.1
A matchmaking result consists of three parts in principle: The description of the initial (1) service request and (2) service o er, and (3) the collection of matching results between pairs of components in both services. The latter express the similarity between individual components in service request and o er, based on their functional and/or semantic description.
We assume that both service request and o er are represented in the
identical form, namely as a complete service description le. Re ecting the variety of
available service description formats for both WS-* and RESTful services, MV
handles the Web Service Description Language (WSDL)2 in conjunction with
Semantic Annotations for WSDL and XML Schema (SAWSDL)3, Web
Ontology Language for Services (OWL-S)4, Web Application Description Language
(WADL) [
Each individual matching result consists of the quali ed names (QNames) of
the two regarded components and a ag that indicates if the two components
have been assigned to each other. The matching result further links to at least one
similarity assessment, which consists of a numerical and textual similarity and a
free-text comment. QNames comprise the namespace of a component as well as
its local name and are unique to each component in a service description. Because
some components, such as individual parameters, may be referenced multiple
times, the QName of any parent of a component may optionally be provided
as well. Both numerical and textual similarity have been included because they
also appear in existing matchmaking approaches. For example, discrete Degrees
of Match are used in [
Because MV constitutes an (uno cial) complement to SME2, we have decided to facilitate SME2's plug-in architecture for the integration of external matchmakers. It requires matchmakers to be provided as a JAR le (optionally with additional libraries) along with an XML-based description. As does SME2, MV requires each compatible matchmaker to implement a Java interface class. This interface, IMatchmaker, speci es one method, matchServices, which conducts the matchmaking process for a service request and o er le. The locations of the respective description les are provided as URLs. The matching result is returned using the Java data structure outlined in the previous section. Alternatively, complete matchmaking results may be provided to MV in the form of XML-based les. This can be useful to matchmakers that do not facilitate MV's plug-in architecture, e.g., if they are not implemented in Java.
MV has been implemented in Java using di erent existing frameworks. For the visualization of graphs, the Java Universal Network/Graph Framework (JUNG)5 is used. To read semantic service descriptions, we employ Woden4SAWSDL6 and OWL-S API7 to read (SA)WSDL and OWL-S documents respectively. For the processing of WADL documents, the generic JDOM8 framework is utilized. hRESTS documents are rst converted into an RDF model, using a XSLT stylesheet by Kopecky9, and subsequently processed using the JENA framework10. 2.4
MV provides a platform-independent Graphical User Interface (GUI) based on the Java Swing framework. A portal site, including a demo video, screenshots, and additional example resources, is available at:
http://www.kom.tu-darmstadt.de/~lampeu/iswc-2010/
The GUI depicts a service request and service o er in the form of a tree structure, representing the increasing level of abstraction. The root, i.e., the whole service, branches into individual leaves, i.e., components, with parameters constituting the most ne-grained level of abstraction. The level of expansion for the tree may be globally selected, thus allowing to focus on the level of abstraction of highest interest. Between all components on the currently selected 5 http://jung.sourceforge.net/ 6 http://lsdis.cs.uga.edu/projects/meteor-s/opensource/woden4sawsdl/ 7 http://on.cs.unibas.ch/owls-api/index.html 8 http://www.jdom.org/ 9 http://cms-wg.sti2.org/TR/d12/v0.1/20081202/xslt/hrests.xslt 10 http://jena.sourceforge.net/ level, the matching results are visualized in the form of edges. By selecting a component, the associated source code from the service description le may be displayed in the GUI. Further, by selecting a connecting edge, details about the respective matching result can be displayed.
MV allows conducting matchmaking on a selected service request and o er using any of the plug-in matchmaking engines. Results may also be combined for direct comparison. That way, users may assess in detail which similarities have been given to pairs of components and why certain components have been assigned to each other. Further, an existing XML result le may be loaded or an individual service may simply be displayed for analysis. 3
In the work at hand, we presented the Matchmaking Visualizer, a tool for the visualization of semantic Web service matchmaking results. It complements existing evaluation tools for semantic matchmakers that rely on rather coarsegrained information retrieval metrics. Matchmaking Visualizer permits a much more ne-grained, in-detail analysis of matchmaking results, down to the level of individual components and o ers support for the direct comparison of di erent matchmaking approaches. For that matter, the tool is compatible with the plugin architecture of the popular SME2 evaluation suite, o ering the convenient integration of external matchmakers.
Acknowledgements. This work is supported in part by E-Finance Lab e. V., Frankfurt am Main, Germany (www.efinancelab.de).