Annotation paths are a new method for semantic annotation which overcomes the limited expressiveness of semantic annotations by concept references as defined in the SAWSDL standard. In this work we show some preliminary evaluation of the feasibility of annotation paths for web service discovery. The experiments suggest that annotation paths can capture the semantics of XML schemas and web service descriptions more precisely and appears as a promising approach for improving web service discovery.
Web service discovery aims at (semi-)automating the search for suitable web
services. A web service discovery systems accepts a service request (a specification
of the needed web service) and a set of web service descriptions (advertisements)
as input and returns a list of web service descriptions ranked by relevance for
the request. There are many different approaches ranging from the structural or
lexical comparison of requests and advertisements to approaches that are based
on the explicit definition of the semantics using ontologies [
Our previous work focused on the annotation of XML-schemas with reference
ontologies in order to automate the generation of executable schema mappings
for document transformations [
To give a first answer on this hypothesis we discusses the usage of annotation paths for web service discovery and report some preliminary results. 2
In some examples we show limitations of simple references to concepts for the
annotation of arbitrary XML-schemas or web service descriptions with existing
reference ontologies. We then present the general concept of annotation paths
(for details we refer to [
The SAWSDL [
The annotated document in Figure 1 exhibits the following problems: { The elements BuyerZipcode and BuyerStreet cannot be annotated because the zip-code is modeled in form of a data-type property and not by a concept in the ontology. { The BuyerCountry element is annotated with the concept country. This does not fully express the semantics because we do not know that the element should contain the country of the buying-party. In addition the SellerCountry element has exactly the same annotation and can therefore not be distinguished. { The attribute Price is annotated with the concept Price. Unfortunately this does not capture the semantics. We do not know the subject of the price (an item) and we do not know the currency.
We have always used exactly one concept for the annotation in the example. However, SAWSDL supports lists of concepts in the modelRef erence attribute but it does not allow to specify the relations between the concepts in this list. Therefore, this does not help to solve the shown problems. In the examples above we have only annotated data-carrying elements. If we would in addition also annotate the parent elements in this case the order element we could add a bit more semantic information. It would be clear that the annotations of the child-elements of the order-element can be seen in the context of an order. Unfortunately this would not help for the ambiguities between the BuyerCountry and the SellerCountry element. In general it would require a strong structural relatedness between the ontology and the annotated XML-Schema or service description which we cannot guarantee when many different schemas or services are annotated with a single reference ontology. In addition SAWSDL does not define that there are any relations between the annotations of parent and child elements. A solution for these non precise annotations is the usage of a more specific reference ontology, which contains concepts that fully match the semantics of each annotated element. For example it would need to contain the concept InvoiceBuyerCountry and InvoiceBuyerZipCode. However, enhancing a general reference ontology with all possible combinations of concepts leads to a combinatorial explosion. 2.2
We propose a new annotation method based on annotation path expressions
that are sequences of steps referring to concepts and properties of a reference
ontology. The first step of an annotation path is always a concept. The last
step of an annotation path can be a concept or a data-type property. Between
two concept steps there is always an object property step. Concept steps can
have constraints denoted in square brackets. In Fig. 3 we give some examples
for annotation paths - more details can be found in [
Annotation paths can automatically be represented in form of OWL2 concepts which can be used to extend the reference ontology. For example the path Order/billTo/Buyer[Mr Smith]/hasCountry/Country is represented as a subclass of Country that has an inverse hasCountry relation to a Buyer whose name is M r: Smith who has an inverse billTo relation to an Order. This can be represented by the OWL expression Country and inv (hasCountry) some (Buyer and fMr Smithg and inv (billTo) some (Order)).
The extraction of annotations from a schema requires to rewrite the schema in order to cope with reused elements first. The resulting schema may contain additional annotations. Since schema elements can refer to other schema elements and types, the full annotation path has to be concatenated from the annotation paths of the elements. For an example an XML-element DeliveryAddress is itself annotated with the annotation path /Order/deliverTo/Address. It has a type definition address. The address type itself contains various elements. One of them is street which is annotated with /Address/hasStreet. In order to construct the complete semantics of the street element which is a child element of DeliveryAddress we get the additional path /Order/deliverTo/Address/hasStreet. 3
To apply the annotation path method to web service discovery we implemented
a logics based service matcher [
The goal of the evaluation is to provide preliminary results whether the
annotation path method leads to better results in service discovery. Therefore, we
have evaluated [
We have annotated a subset of the SAWSDL-TC31 data-set with our annotation path method and have evaluated our matcher against service matchers that took part in the International Semantic Service Selection Contests 2. We have evaluated two scenarios: { Scenario 1: The goal of this scenario was to evaluate how, our simple matcher can compete against current state of the art matchers based on existing requests and advertisements of the SAWSDL-TC-3 data-set. In the first scenario we have selected one arbitrary request (book-price) and 40 advertisements and have annotated them manually using the annotation path method. Our matcher operated on requests and advertisements which are annotated with annotation-paths and the reference matchers used the original annotations and advertisements of the TC-3 data-set. { Scenario 2: The goal of the second scenario was to asses how our matcher competes against other matchers if the semantics cannot be expressed by simple concept annotations. In this case matchers operating on simple concept annotations can only infer the missing semantics by exploiting other dimensions such as the structure or naming of elements. The second scenario was also evaluated using existing advertisements of the SAWSDL-TC3 data-set. We have only changed the request. We now require for the price of books in Euro but excluding tax and we restrict the input to science fiction comics. This cannot be expressed with the used ontologies because no such concepts exists. However, a hint for the standard SASWSDL matchers was provided by the requested output type (EuroPriceExcludingVAT ) and input type (ScienceFictionComic).
We have executed the evaluation with the Service Matchmaker and Execution
Environment (SME23) which is also used for the International Semantic Service
Selection Contests. Due to the partial TC3 data-set we were not able to execute
all matchers. However, we could execute two major representatives iSem[
While these preliminary results do not yet allow to draw final conclusions the annotation paths approach is promising for improving web service discovery. Our simple path-based matcher could clearly show its advantage in Scenario 2 and in Scenario 1 it could compete well with existing state of the art matchers which use far more advanced matching methods and additional aspects of service descriptions and advertisements. 5
The annotation path method for semantic annotation has been developed to overcome limitations in the expressiveness of simple concept references. We showed in some feasibility tests that already a simple implementation of an
annotation path based XML-schema matcher used for comparing web service advertisements with service requests can successfully compete with state-of the art web service discovery systems. We therefore conclude that annotation paths are well suited for capturing the semantics of objects in much finer detail and that the annotation path method and matchers based on it are promising approaches to improve web service discovery. Encouraged by the promising results we plan to evaluate our annotation path based matcher with a larger data-set and against additional existing matchers. Other future work is to integrate our matcher into existing state of the art matchers to gain even better results. Another direction of future work is to evaluate not only the matching precision but also the minimum amount of manual work to semi-automatically create annotation path annotations in comparison to simple concept references.