SEMANTIC Web Services (SWS) [1], like conventional web

Transcription

1 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 25, NO. 5, MAY Toward SWSs Discovery: Mapping from WSDL to OWL-S Based on Ontology Search and Standardization Engine Tamer Ahmed Farrag, Ahmed Ibrahim Saleh, and Hesham Arafat Ali Abstract Semantic Web Services (SWSs) represent the most recent and revolutionary technology developed for machine-tomachine interaction on the web 3.0. As for the conventional web services, the problem of discovering and selecting the most suitable web service represents a challenge for SWSs to be widely used. In this paper, we propose a mapping algorithm that facilitates the redefinition of the conventional web services annotations (i.e., WSDL) using semantic annotations (i.e., OWL-S). This algorithm will be a part of a new discovery mechanism that relies on the semantic annotations of the web services to perform its task. The local ontology repository and ontology search and standardization engine are the backbone of this algorithm. Both of them target to define any data type in the system using a standard ontology-based concept. The originality of the proposed mapping algorithm is its applicability and consideration of the standardization problem. The proposed algorithm is implemented and its components are validated using some test collections and real examples. An experimental test of the proposed techniques is reported, showing the impact of the proposed algorithm in decreasing the time and the effort of the mapping process. Moreover, the experimental results promises that the proposed algorithm will have a positive impact on the discovery process as a whole. Index Terms Semantic Web Service, ontology, mapping, WSDL, OWL-S, ontology-based standardization Ç 1 INTRODUCTION SEMANTIC Web Services (SWS) [1], like conventional web services, are the server end of a client-server system for machine-to-machine interaction via the World Wide Web. The main idea of the Semantic Web [2] is to extend the current human-readable web by encoding some of the semantics of resources in a machine-processable form. Moving beyond syntax opens the door to more advanced applications and functionalities on the web. Computers will be able to search, process, integrate, and present the content of these resources in a meaningful manner. As a Semantic Web component, SWSs use markups that make the data machine-readable in a detailed and sophisticated way (as compared to human-readable HTML pages which are not usually easily understood by computer programs). As shown in Fig. 1, there are five steps to deal with any web service (WS) [3]: 1. the process of Advertisement in which the service provider aims to publish information about the benefits of the service and how to use it;. T.A. Farrag is with the Department of Computers and Systems, Faculty of Engineering, Mansoura University, PO Box 450, Mansoura, Dakahlia, Egypt. tfarrag2000@hotmail.com.. A.I. Saleh and H.A. Ali are with the Department of Computers and Systems, Faculty of Engineering, Mansoura University, Mansoura, Dakahlia 35516, Egypt. aisaleh@yahoo.com, h_arafat_ali@mans.edu.eg. Manuscript received 3 Feb. 2011; revised 24 Sept. 2011; accepted 20 Jan. 2012; published online 13 Feb Recommended for acceptance by T. Grust. For information on obtaining reprints of this article, please send to: tkde@computer.org, and reference IEEECS Log Number TKDE Digital Object Identifier no /TKDE the process of Discovery which can be defined as the process of finding the list of services that can possibly satisfy the user requirements; 3. the process of Selection that targets to select the most suitable WS. This process is usually based on application-dependent metrics (e.g., QoS); 4. the Composition process that integrates the selected WSs into a compound process; 5. the process of Invocation that invokes a single WS or compound process by providing it with all the necessary inputs for its execution. An extended definition of the discovery process is mentioned in [4], which states that The discovery process is the process of finding a service that can possibly satisfy user requirements, choosing between several services, and composing services to form a single service. In [5], we introduce our new discovery mechanism. This mechanism has the ability to provide optimal results for any service request. This mechanism is distinguished by its service repository, which is built using the advertisements of semantic and non-sws. In addition, its method of organizing and classifying the available service advertisements improves the speed and quality of the discovery process. Fig. 2 shows the discovery mechanism system organization. We can summarize our mechanism into two phases: Phase 1 Database Creation : in this phase the database of the discovery mechanism is created using the semantic definition of the registered services, it consists of three main steps: 1. Mapping from the WSDL [6] of the already existing WSs to a semantic definition using OWL-S [7], [8] /13/$31.00 ß 2013 IEEE Published by the IEEE Computer Society

2 1136 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 25, NO. 5, MAY 2013 Fig. 1. Web services life cycle. 2. Registering all the available semantic definitions (whether they belong to WSs or SWSs) in the Unclassified Profiles database. 3. Classifying these data into preprepared clusters to make the discovery easier and faster. Phase 2 Discovery Process : this phase starts with receiving the user request of a certain service and then follows these steps: 1. Use the discovery algorithm to search into the database for the suitable results. 2. Use the ranking algorithm to rank the results to enhance the user selections. This paper introduces a mapping algorithm which represents the first step in the first phase of the proposed discovery mechanism. This algorithm aims to redefine the conventional web services using semantic markups. This does not only mean the process of converting the conventional web service description language (WSDL) to a semantic one (i.e., OWL-S), but it also means the standardization of this definition by using the concept of ontology to describe any type of data in the service. Consequently, the proposed algorithm contains an important component called the ontology search and standardization engine (OSSE) that helps in the standardization process. OSSE s function is based on searching for a suitable ontology in the local ontology repository. This paper is organized as follows: Section 2 introduces a background about WSDL and OWL-S. Section 3 introduces the related work. Section 4 introduces the local ontologies repository. Section 5 introduces the OSSE. Section 6 introduces the mapping algorithm. Section 7 introduces examples on how to apply the mapping algorithm. Section 8 concludes this paper. 2 BACKGROUND In this section, an overview of the WSDL and its basic components is presented. WSDL represents a nonsemantic definition language; so, this paper introduces algorithms used to redefine WSDL using one of the semantic definition languages. Then, an overview of the web ontology language for services (OWL-S) is presented. OWL-S is one of the most important and widely used semantic definition languages for web services. There is many background information which is cited within the different sections as the context needed. Fig. 2. Discovery mechanism system organization. 2.1 Web Service Definition Language WSDL [6] is an XML [9] format used to describe network services as a set of endpoints operating on messages containing either document-oriented or procedure-oriented information. WSDL is often used in combination with SOAP [10] and an XML Schema [11] to provide web services over the Internet. A client program connecting to a web service can read the WSDL to determine what operations are available on the server. Any special data types used are embedded in the WSDL file in the form of XML Schema. The client can then use SOAP to actually call one of the operations listed in the WSDL. In 2007, W3C 1 introduces version 2.0 of WSDL [12]. WSDL 2.0 has many new features such as interface inheritance, extensible message exchange patterns, and an abstract component model. Fig. 3 summarizes the basic components of the WSDL file where:. Service: The service can be thought of as a container for a set of system functions that have been exposed to the web based protocols.. Port/Endpoint: The port does nothing rather than defining the address or connection point to a web service. It is typically represented by a simple http 1. The World Wide Web Consortium.

3 FARRAG ET AL.: TOWARD SWSS DISCOVERY: MAPPING FROM WSDL TO OWL-S BASED ON ONTOLOGY SEARCH AND Fig. 3. WSDL basic component. URL string. It has been renamed to <endpoint> in WSDL Binding: Specifies the interface, defines the SOAP binding style (RPC/Document) and transport Protocol (SOAP). The binding section also defines the operations.. PortType/Interface: The <porttype> element, which has been renamed to <interface> in WSDL 2.0, defines a web service, the operations that can be performed, and the messages that are used to perform the operation.. Operation: Each operation can be compared to a method or function call in a traditional programming language.. Types: The purpose of the types in WSDL is to describe the data. XML Schema is used (inline or referenced) for this purpose. 2.2 Web Ontology Language for Services OWL-S [7] defines an upper ontology used to describe the properties and capabilities of web services in OWL. OWL-S is a W3C submission since 2004 and from that day on many researchers use it to describe SWSs [13], [14]. The OWL-S authors target to enable automatic web services discovery, invocation, composition, and interoperation. Fig. 4 shows the basic components of the service class used to describe the web services using OWL-S. There are three main components: 1. The service profile that describes the function of the service and provides the all necessary information that helps in the discovery process and answer the question What does this service do? 2. The service model that describes all the processes the service is composed of, how these processes are executed, under which conditions they are executed and answer the question How does this service work? 3. The service grounding that plays the role of coordinator of the service usage. Therefore, it is responsible for the protocols and mapping with traditional web service standards such as WSDL and SOAP. In this study, we care about the process of discovery, so we will present the OWL-S service profile component in more details because its information will be the base of our discovery mechanism. Fig. 5 illustrates the main information represented by the service profile. Service profile Fig. 4. Top level of the service ontology. information can be categorized into three categories: 1) information that can be readable by human beings, it includes the service name, a description that presents the function of the service and the contact information of the service provider; 2) information that deals with the functionality of the service. This information includes full data about Inputs, Outputs, Preconditions and Effects (IOPE). The OWL-S Profile represents two aspects of the functionality of the service: the information transformation (represented by inputs and outputs) and the state change produced by the execution of the service (represented by preconditions and effects); 3) information that includes the quality guarantees, called profile attributes, that are provided by the service, possible classification of the service, and additional parameters that the service may want to specify. 3 RELATED WORK There has been a little effort spent on transforming between WSDL and OWL-S. Generally, it is noticed that this area of research did not receive sufficient attention. In [15], the authors introduce a mapping tool called Automated Semantic Service Annotation with Machine Learning (ASSAM). ASSAM generates OWL-S file from WSDL file. This tool suffers from the following limitations: 1. The tool does not provide an organization for the available ontologies. This lack of organization yields Fig. 5. Service profile information.

4 1138 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 25, NO. 5, MAY 2013 in many problems specially when the tool is used in a real SWSs system where the existence of a huge number of ontologies and concepts (classes) is expected. 2. The tool introduces a list of different choices to the user to select the most appropriate class that can represent a semantic definition for each data type in the WSDL file. This list is an unordered (unranked) list. So, the choosing process is difficult for any user. 3. The process of finding the possible list of classes for any data type is just a text-based search process. 4. Most of topics that are discussed and evaluated in this research concentrate on the process of web services classification. So, the main contribution of the research is in the field of classification. In other words, ASSAM can be considered as a novel web services classification tool instead of mapping tool as the authors claims. MWSAF is a web service annotation framework which is introduced in [16] and modified in [17]. It annotates WSDL descriptions of the services with metadata from relevant ontologies. Due to the difference in expressiveness of WSDL and OWL, it is difficult to directly match the two formats. MWSAF proposes to bridge the gap by converting each of them to a common representation called schema- Graphs. A schemagraph is simply a graph or tree representation of the XML or OWL-S document. The conversion rules are explained in-depth in [16]. Once the schemagraphs have been created, matching algorithms are executed on the graphs in order to determine similarities. Once a concept is matched against all the concepts in an ontology, the concept that has the highest matching score is the best mapping choice. Several algorithms have been defined for matching. Although the authors introduce a novel tool, we have some comments: 1. One of the most important components of the tool is the Ontology store which stores the ontologies used by the system to annotate the web service. The insertion methodology is not represented in details but as we infer, the new ontologies are categorized into domains without collecting information about its internal structure. 2. As in ASSAM, the evaluation process concentrates on measuring the tool capability to make a correct classification decision. However, neither evaluation nor case study about mapping process is provided. 3. In addition, the evaluation process tries to prove that as the ontology concepts definition improves, the matching score increases. We think that it is a fact that does not need to be proved. In [18], the authors introduce a tool called WSDL2OWL- S. WSDL2OWL-S provides a transition between WSDL and OWL-S. The results of this transformation are a complete specification of the Grounding and a partial specification of the Process Model and Profile of OWL-S. The incompleteness of the specification is due to the fact that OWL-S and WSDL differs in the information contained. The output of WSDL2OWL-S converter provides the basic structure of OWL-S description of the web services and saves a great deal of man power. However, this approach suffers from several limitations: 1. The authors state that they choose to force the translation process of XSD complex type to generate concepts which are totally unrelated to the ontologies available in the Semantic Web. This choice will cause a generation of a huge number of concepts when dealing with the large number of web services that are expected to be registered in the system. We can title this problem as lacking of standardization problem. We solve this problem by proposing OSSE and the local ontologies repository. These are essential components of the proposed mapping algorithm that aims to solve the lacking of standardization problem. 2. This tool has been designed in isolation from the discovery mechanism design. On contrary, our proposal is a part of a new proposed discovery mechanism. So, the mapping process collects information and organizes it in a way that will ease the discovery process. 4 LOCAL ONTOLOGIES REPOSITORY The main challenge that prevents the Semantic Web from achieving its goals is the problem of standardization. The key point of any standardization research is how to use the ontology concept to define different data types in the system. Most of the researches focus on how to write this ontology. In 2004, Web Ontology Language (OWL [19]) became the W3C official language for ontology writing. Also, many projects focus on how to collect the largest number of ontologies to construct some ontologies libraries. Examples can be found at DAML Ontology Library website, 2 Sem- WebCentral, 3 and with the Protégé ontology editor. Swoogle 4 can be considered as the best repository of online ontologies as it has cataloged over 10,000 Semantic Web documents and has indexed metadata about their classes, properties, and the relationships among them. In general, there are many drawbacks of these attempts such as: 1) the absence of any clear methodology of how to insert a new ontology into the library or how to update the structure of the already existing ontologies; 2) the absence of any searchable metadata about the classes, properties, and relations included in the ontology (Swoogle is excluded from this drawback); 3) the absence of homogeneity among ontologies as they are written in different languages and also the lack of metadata about ontology-to-ontology relationships. The Local Ontologies Repository is constructed to avoid these drawbacks that may reflect on the discovery process in a negative manner. In the following sections, we introduce the following: 1) the repository logical structure; 2) The methodology of inserting a new ontology into our repository; 3) The implementation and statistics of the repository

5 FARRAG ET AL.: TOWARD SWSS DISCOVERY: MAPPING FROM WSDL TO OWL-S BASED ON ONTOLOGY SEARCH AND Repository Logical Structure Our repository is built based on the object-oriented paradigm (OOP) which can be considered as one of the most remarkable approaches to describe the ontologies relationships, as will be shown in this section. Because of the acknowledged success and wide spread of using the OOP in the field of programming, many researchers try to import it to the Semantic Web. There are many researches in this field, [20], [21] describe how to introduce similarity measures inspired from the domain of OOP for ontology concept matching. In [22], the author attempts to adapt existing object-oriented software development methodologies to the task of ontology development, selected approaches are originated from research in artificial intelligence, knowledge representation, and object modeling are presented. In [23], an approach to object-ontology mapping is presented and JAVA is used for implementation and validation. As in OOP, OWL ontology classes and properties are defined as instances of appropriate built-in classes (e.g., owl: Class or owl: ObjectProperty). So, the model of OWL ontologies is very similar to OOP. Any concept in a domain should correspond to a class that belongs to one of the taxonomic trees. Every individual in OWL is a member of the class owl:thing that corresponds to the Object class in OOP. Thus, each user-defined class is implicitly a subclass of owl: Thing. In [24], the authors provide a detailed review on the differences and similarities between Semantic Web languages and object-oriented languages. In addition to the OOP success in many fields, the similarities between the OWL ontology concepts and OOP concepts encourage us to accredit OOP as the logical structure of our repository. Some information, about every ontology in the repository, will be collected and updated continuously. This information includes the parent of the ontology class (super class), number of properties, how many times this ontology appeared in the mapping process results as will be illustrated in the next sections. 4.2 Inserting New Ontology Methodology The main target of inserting a new ontology to our repository process is to collect two types of data: 1) Data concerning the logical structure, 2) Keywords relevant to the new ontology. These data are used by the OSSE to respond to the requests for the concepts. The first type of data are used by the structural refining stage of OSSE; while the second type is used by the linguistic search stage of OSSE. The OSSE will be discussed next in this paper. After collecting the required data, they are saved into the repository file and database system. As shown in Fig. 6, two concurrent processes collect these two types of data: the first process is called Structure Extraction. During this process, data concerning ontology concepts (classes), properties, and relationships are collected. Despite the high importance of ontology instances, axioms and many other aspects of the ontology, the data concerning them are not important for the discovery process (the main target of our research). That does not mean to lose these information but all the ontology contents are stored in the repository file system that can be considered as the original source of information for any ontology. The database is a mean to speed up the process of finding the required ontology information for the SWSs discovery process. Fig. 6. Insert New Ontology to "Local Ontology Repository." The second process is called Keywords Extraction where text-mining techniques will be used to extract the keywords form the whole OWL file. In this process, we deal with the OWL file as a text file that contains information about the inserted SWS. There is no doubt that processing the whole file may be time consuming but it provides a clearer vision than processing specific portions of it. This process consists of five steps: 1. Tokenization: is the process of splitting the text into very simple tokens such as numbers, punctuation, and words of different types. In this context, we concentrate on word tokens only. 2. Lemmatization: is the process of reducing inflectional forms and sometimes derivationally related forms of a word to a common base form. For instance: am, are, is ) be, cars, car s, cars ) car, best, better ) good. We use WordNet [25] to perform this process. 3. Now, we have a long list of words. So, we remove the extremely common words which are called stop words such as (a, an, am, will, he, she, their ;..., etc). We depend on Google 5 list 6 which contains 659 words. 4. Then, we remove any word that appears in the keywords list of the languages used to write the ontology such as (XSD [11], OWL[19], RDF[26], RDFS [27];..., etc). 5. Finally, we compute the term frequency (TF) for each word in the list. To demonstrate how to calculate TF, consider a document containing 100 words wherein the word cow appears three times. Then, the TF for cow is then ð3=100þ ¼0:

6 1140 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 25, NO. 5, MAY 2013 Fig. 7. Inserting new ontology GUI snapshot. 4.3 Repository Implementation and Statistics To construct our repository, we use the OWLS-TC v3.0 collection [28] which contains 1,007 indexed OWL-S services and 23 different ontologies. These ontologies are used to describe the semantics of I/O concepts of services in this test collection. Fig. 7 shows a snapshot of the GUI of a tool that is used to insert a new ontology into our repository according to the previously described methodology. This tool is not available online until now. It will be available as a part of a SWSs discovery tool based on our proposed mechanism. Fig. 8 illustrates the database design that makes the process of finding information about repository contents faster and easier. After inserting all the 23 ontologies, our repository contains 3,503 concepts, 1,145 properties, 5,211 relationships, and 5,002 keywords. The important question that should be answered is why do we use the database approach instead of full ontology approach? The answer is that the main role of ontology is to provide the ability to define all the system data types in a semantic manner. The database is not an alternative for ontology. It is used to assist the process of finding the suitable ontology. The other replacement of the database approach is using the plain text files searching techniques which is certainly a very slow and inefficient replacement. 5 ONTOLOGY SEARCH AND STANDARDIZATION ENGINE The main challenge that faces the development of Semantic Web is the problem of standardization. Therefore, there are two points to concern about: 1) it does not make any sense to find the same definition for a certain concept repeated more than one time in the system. This does not include the case when there are different definitions for the same concept and each of them adds new information. For instance, one ontology might define the class Animal equivalent to Living thing AND NOT Human. Another ontology might define the class Animal as the Union Of Herbivores AND Carnivores. Both definitions are correct, since they highlight different aspects or properties for the class animal, and they cannot be considered mutually exclusive. 2) It is illogic to define a concept isolated from the already existing concepts definitions. In other words, it is Fig. 8. Inserting new ontology GUI snapshot. better from standardization point of view to try to find a relation between the new concept definition and the already existing ones. Fig. 9 presents the OSSE. This engine takes in consideration the two points mentioned before. The input is a required concept with certain features (properties) and the output is an ordered list of possible related ontologies. The concept request is formalized in form of a temporary OWL ontology that contains one concept. The name and the structure of the required concept is the primary supplied information used by OSSE. The engine has three main stages: linguistic search, structural refining, and statistical refining. These three stages are described below: 1. Linguistic search. In this type of search, OSSE uses the text mining techniques same as used in Section 4.2 to extract the most important keywords in the concept request. Then, it tries to find all the synonyms and related words to expand our list by the aid of WordNet [25], which includes over 90,000 word senses. For example, if the system needs to find car concept, it should search also for many synonyms and related words (e.g., auto, automobile, machine, motorcar, four-wheel drive, 4WD, and railcar). This list of keywords (and its related words) is used to search in the local ontologies repository to get a list of possible related ontologies. The relevant keywords, which are collected using the inserting methodology presented in the previous section, are considered to be the base of the linguistic search. The list of ontologies is arranged based on the keywords that they contain in terms of term frequency. For each ontology, the summation of term frequency values of each keyword that belongs to the concept request keywords list and belongs to the ontology at the same time is computed. This summation represents a measure of the degree of the ontology linguistic relevance (OLR). Degree of OLR can be calculated by OLR ¼ i¼nck X i¼0 gett Fðk i Þ; ð1þ

7 FARRAG ET AL.: TOWARD SWSS DISCOVERY: MAPPING FROM WSDL TO OWL-S BASED ON ONTOLOGY SEARCH AND Fig. 9. Overview of the OSSE. where NCK is the total number of concept request keywords. The function gettf() is used to get the previously computed TF value of the keyword (K) in the tested ontology. For remember, the TF value is computed for each ontology keyword during the process of new ontology insertion that is discussed in Section 4.2 of this paper. Due to the expected large number of available ontologies, the system administrators can set a threshold value for OLR. In some cases, OSSE fails to find any related ontology in the local repository. Therefore, it asks Swoogle for help to find some OWL related ontologies. If there are results for this search, OSSE downloads the top five ontologies (or any number of ontologies chosen by the administrator). If the service provider accepts any of these downloaded ontologies, the system automatically inserts this ontology to the local ontology repository using the inserting methodology. This process grantees that our ontology repository is extended to satisfy the service providers needs without changing the features of our repository. Downloading and inserting the selected ontology in the repository is a process that consumes undetermined time. This time depends on the downloading speed and the degree of ontology complexity. It is important to note that the inserting process is performed after finishing the process of choosing the suitable concept. Therefore, this process does not affect the performance of OSSE and it can be done as a background process. In some rare cases, searching in Swoogle returns with no results. In this case, OSSE inserts the temporary ontology into the local repository using the inserting methodology discussed in Section 4.2. A long list of related ontologies is expected. Therefore, the next two steps aim to reduce and rank this list. 2. Structural refining. In this stage, OSSE refines the list produced by the linguistic search. This refining is performed by searching in each ontology in the list to find any concept related to the required concept. If OSSE does not find any related concept in a particular ontology, this ontology is deleted from the possible ontologies list. Data concerning the logical structure which are collected using the inserting methodology, are considered to be the base of the structural refining. We present a simple example to clarify what Related ontology means, a simple example is designed to clear up that expression. We assume that the requested concept is called C R and it has three properties {P1, P2, and P3}. C R is compared against another concept called C which is part of four suggested concepts-trees that represent all the possible concepts structure in any ontology. Fig. 10 illustrates the four possible alternatives of Concept-toconcept Relationships: 1. identical relation: C R and C have the same properties; 2. super relation: C may be considered as a parent of C R ; 3. sub relation: C may be considered as a child of C R ; 4. neighbor relation: C and C R have some common properties.

8 1142 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 25, NO. 5, MAY 2013 Fig. 10. Concept-to-concept relationships. Back to the process of structural refining, the ontology is checked to answer four serial questions that correspond to the four possible concept-to-concept relationship as shown in Fig. 9. When we have a yes answer to any question of them, OSSE stops the checking process of the ontology and assigns a rank value for it. Identical relation has the highest ranking value followed by Super, Sub, and Neighbor relations in order from highest to lowest ranking value. After completing the process of checking the ontologies list, OSSE reorders the ontologies list according to the computed ranking value. 3. Statistical refining. The choices of the service provider are stored in the Concepts Mapping History database. These data are used by OSSE to rerank the possible ontologies list. If there are two ontologies with the same rank in the previous step, OSSE uses this historical data to know the most preferred ontologies for the services providers. After these three steps, OSSE has an ordered possible ontologies list for the concept request. This list is presented to the service provider, who chooses the most suitable ontology from his point of view. His choice recorded in the Concepts Mapping History database and the selected ontology edited automatically, if it is needed as shown in Fig. 11. The process of ontology editing causes some changes in the database and the file system of the local ontologies repository. So, we design a simple C#.NET tool to modify the ontology OWL-S file to apply the new changes in the ontology logical structure. This tool is not a stand-alone tool but it is one of many sub tools that are used to implement our proposed mapping algorithm. 5.1 OSSE Implementation OSSE depends on the database of the local ontology repository to retrieve any required information about the already exiting ontologies. This dependency makes OSSE Fig. 11. Ontology concepts-trees after editing. work in fast and accurate manner. OSSE has three main features that should be tested. The first one is its capability to find the matched ontology for an already existing concept. Second, its ability to access Swoogle web service to download suitable ontologies for a requested concept that does not belong to the local repository. Third, its ability to extract the suitable concept-to-concept relationships. Three experiments are designed to test these features: Experiment 1: Goal: Test the linguistic search step of OSSE. Procedure: 1. Request for Car concept. 2. OSSE extracts keywords ) car 3. Extend the keywords using WordNet ) auto, automobile, machine, motorcar, gondola, railcar 3. Search local ontology Repository. In details, OSSE searches into the repository database, i.e., keywords table. Results: The search yields a list of six possible related ontologies. Experiment 2: Goal: Test the usage of external search, i.e., Swoogle web service to download required ontologies. Preparations: before starting this experiment we delete the six related ontologies for the concept Car Procedure: 1. Request for Car concept. 2. OSSE extracts keywords ) car 3. Extend the keywords using WordNet ) auto, automobile, machine, motorcar, gondola, railcar 4. Search local ontology Repository. In details, OSSE searches into the repository database Results: The search yields no results. So, OSSE uses Swoogle to down load the top five ontologies (i.e., the URL of the top ontology is owl ).

9 FARRAG ET AL.: TOWARD SWSS DISCOVERY: MAPPING FROM WSDL TO OWL-S BASED ON ONTOLOGY SEARCH AND Fig. 12. Mapping algorithm abstract view. Experiment 3: Goal: Test the ability of OSSE to extract the suitable conceptto-concept relationships. Procedure: 1. Request for Address concept has properties (country, city, street, and building). 2. OSSE extracts keywords ) address, country, street, and building. 3. Extend the keywords using WordNet ) address, destination, country, state, nation, area, city, metropolis, urban, street, building, edifice, construction. 4. Search local ontology Repository. In details, OSSE searches into the repository database 5. Perform the structural refining Results:. OSSE linguistic search yields a list to 19 possible related ontologies.. After structural refining this list contain only 7 ranked ontologies and the top ranked ontology is (portal. owl). By manual check this ontology has a concept called address and has 9 properties (Area, building, city-or-village, country, number, postcode, pretty-label, region, street) 6 MAPPING ALGORITHM Almost all existing web services are described using WSDL, which is a nonsemantic definition language. To benefit from the advantages of Semantic Web systems, the service provider needs to redefine his service using semantic annotations. This redefinition can occur by mapping from WDSL to a semantic definition language (e.g., OWL-s). The problems of wasted time, nonaccurate mapping and absence of any standardization may be the most important issues that the service provider concerns about. Our Fig. 13. Proposed mapping algorithm. proposed algorithm helps the service provider to redefine his service using OWL-S. No one can propose a fully automatic algorithm because WSDL, by nature, lacks many information that is needed to have a complete semantic definition. Therefore, the proposed algorithm can be considered as a semiautomated tool to help the service provider. During the mapping process, WSDL types are rewritten using OWL and standardized using the OSSE component. In the future, we target to deal with other semantic definition languages and frameworks such as: WSMO [29], WSDL-S [30], SAWSDL [31], SWSF [32]. 6.1 Mapping Abstract View Fig. 12 represents the roadmap of our proposed algorithm, it depicts the main components of WSDL and OWL-S files and shows the relations between them. The figure also shows which information can be extracted automatically by the system, which should be supplied by the service provider and which is out of our scope. As stated before, OWL-S concentrates on the description of functional properties of SWS. But, it cannot be used to describe the QoS and other nonfunctional properties. In [33], OWL-Q is presented as an extension to OWL-S which will help the service provider to describe the nonfunctional properties of his service. So, we add a special section in the service profile for any additional QoS parameters supplied by the service provider. 6.2 The Proposed Mapping Algorithm Fig. 13 depicts the proposed mapping algorithm. The algorithm has two phases, the first one is called the automated phase in which the WSDL file is parsed to extract information to fill the required properties to construct a service profile in the format of OWL-S. As WSDL does not contain all the required data especially the nonfunctional one, the second phase, called the manual phase, is designed in a way that the service provider can interpolate the required data. The automated phase consists of three steps:

10 1144 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 25, NO. 5, MAY The system recommends ontology for each XSD type in the WSDL file. This process is based on our Local Ontologies Repository. The details of this process are presented in Section The system extracts the service name and its text description provided by the service provider. 3. The system extracts the inputs and the outputs of the service. In manual phase, the service provider will be asked by the system to: 1. Define the preconditions and the effects of the service if found. Define the nonfunctional properties of the services such as (cost, response time, location;..., etc). A tool that applies this mapping algorithm is implemented. This tool accepts the WSDL file then it applies the automated phase of the mapping process then the results of this phase are presented to the service provider. Using the GUI interface of this tool, the service provider reviews the results and chooses the most appropriate ontology concept for each data type. The list of the possible corresponding concepts for each data type is prepared by the Types Converter component. 6.3 WSDL Type to OWL Ontology Converter (Types Converter) One of the most important steps in the process of mapping WSDL to OWL-S is the conversion from the WSDL types, typically XSD types, to OWL ontologies. There are two categories of XSD types: 1) Primitive (simple) XSD types, e.g., string, integer that need to be converted to OWL ontologies. They are defined directly as inputs or outputs of an atomic process in the service model file; 2) Complex XSD types which are translated into OWL ontology concepts whose properties correspond to the elements in the translated type. According to [15], [18], the converter of complex XSD types has two alternative designs. The first one is to generate OWL-S specifications that make use of XSD types and make no use of OWL ontologies. The second alternative is to generate concepts that could be totally unrelated to ontologies available in the Semantic Web and therefore they are definitely useless from the automatic reasoning point of view. From our point of view, there is a third alternative that is to search into the Local Ontology Repository using the OSSE component to find a ranked list of the most related ontologies that already exist. Then, ask the service provider to choose the most suitable ontology concept. If the service provider does not accept any of the already existing ontologies, the system uses the second alternative. Fig. 14 presents the steps of the conversion process. The XSD types are extracted one by one from the WSDL file. The first question is Is this primitive or complex type? If it is primitive the conversion process does not work and keeps it as it is. On the other hand, if it is a complex type the converter starts a process to find the most suitable OWL ontology. The process starts by creating a temporary ontology where each XSD complex type can be described by an ontology has one concept and many properties. This concept represents the input of the OSSE component. The output of OSSE is a ranked related ontologies list. This list Fig. 14. Types converter. is offered to the service provider to choose the most suitable ontology. In some rare cases, OSSE fails to find any related ontologies. In this case, the system inserts the temporary ontology to the local ontologies repository using the insertion methodology shown in Section RESULTS AND MAPPING EXAMPLES The proposed mapping algorithm is implemented. For testing purposes, we use many WSDL files as a case study to monitor the mapping process (specially the automatic phase) and to evaluate its results. Some of these files belong to OWLS-TC [28] which provides the WSDL file and its corresponding OWLS file. This collection is used to compare the results of our mapping process with those already included within the collection. The rest of the WSDL files belong to real web services (e.g., Yahoo! Mail web service) which help us to study the behavior of our algorithm in a practical environment. Next, we discuss the results of the mapping process and present one example from each group of WSDL files. 7.1 OWLS-TC Validating the proposed algorithm is done by using 1,000 of WSDL files which are included in OWLS-TC v3.0 and mapped using the proposed algorithm. The automatic phase of the mapping process is done within 1 hour. This time can be neglected when compared with the time consumed in the case of manual mapping. Fig. 15 shows a comparison between the WSDL2OWL-S tool [18] and the proposed mapping algorithm. The figure presents the relation between the number of concepts generated by the system and the number of registered services. We can obviously note that the number of used concepts will be increased when the number of services increases in both cases. But, in the case of WSDL2OWL-S

11 FARRAG ET AL.: TOWARD SWSS DISCOVERY: MAPPING FROM WSDL TO OWL-S BASED ON ONTOLOGY SEARCH AND Fig. 15. No. of used concepts versus no. of registered services. tool the number of concepts increase with very high rate when compared to the case of our proposed algorithm. For example, when the number of registered services becomes 1,000, the average number of concepts per service is 2.66 in case of WSDL2OWL-S and 0.4 in case of the proposed Fig. 17. Output of mapping for BookPriceTypeSize Service. algorithm. So, the proposed algorithm is more scalable than WSDL2OWL-S. It is important to state that the performance of the discovery process is negatively affected when the number of concepts defined in the system increases. That is because the study of inputs/outputs matching between the request and the available services is a major task for any discovery process. There is no doubt that this matching become faster and more accurate when the total number of concepts which defines the inputs and outputs types become smaller. So, the proposed algorithm will be better than WSDL2OWL-S from the discovery point of view. As stated in the related work section, the authors of ASSMA [15] and MWSAF [16], [17] did not provide any evaluation data concerning the mapping process. The data provided about the experiments and the evolution concentrates on the classification process which is out of our current scope. Now, we present one of these WSDL files (BookPricesizebook-type.wsdl) which describes a service called BookPriceTypeSizeService as shown in Fig. 16. The function of this service can be understood from its provider description. He says: This service informs you with not only an affordable price of the book but also its type (hard-cover or paper-back) and size (small/large/medium). So, this service contains one operation that has one input (a book) and three outputs (price, book-type, size). Fig. 17 shows the output of our mapping algorithm (owls file) for this service. When comparing between Figs. 16 and 17, it is noticed that the service inputs and outputs data types are extracted and redefined by types from the local ontology repository. For example, the input of the type Book in the WSDL file is redefined as This redefinition is a result of the mapping standardization process. 7.2 Yahoo! Mail Web Services The Yahoo!7 Mail web service API8 is a full-featured interface to Yahoo! Mail. With it, you can build applications that display message summary information, parse message contents, manage folders, and even compose and send Fig. 16. WSDL file of BookPriceTypeSize service

12 1146 IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 25, NO. 5, MAY 2013 messages. The WSDL description (ymws.wsdl) of the Yahoo! Mail web service contains 21 operations to perform the tasks described before, 121 types, and 42 input/output messages (there is no use of fault messages). As stated before, the mapping process consists of two phases the automated phase and the manual phase. In the case of mapping the Yahoo! Mail web service, the automated phase generates the corresponding OWL-S file and a list of the most related ontologies corresponded to each service data type. This phase takes less than 1 minute. In addition, the manual phase that is performed by the service provider to choose the most appropriate ontology for each data type, takes around 1 hour. In the absence of such a mapping algorithm, the translation of a complex WSDL document (such as the specification of the Yahoo! Mail web service) takes about a week of man time. Therefore, there is a significant saving in the time spent. 8 CONCLUSION AND FUTURE DIRECTIONS In this paper, we target to solve the problem of enabling web services discovery. We propose a mapping algorithm that helps to facilitate the integration of the current conventional web services into the new environment of the Semantic Web. This has been achieved by extracting information from WSDL files and using it to create a new semantic description files using OWL-S. The proposed algorithm contains a basic component called Types Converter which is used to convert XSD complex types to ontologies. This converter depends on OSSE component that uses the Local Ontology Repository to find a suitable ontology for each XSD complex type. The bottom-up design approach is used to implement and validate our mapping algorithm. So, we start with creating the local ontology repository then we implement the OSSE component and finally we implement the type converter and the mapping algorithm. Many designed and real examples are used to verify the applicability of the proposed algorithm. The results prove that this algorithm helps the service providers to cut down the effort and the time consumed in redefining their services in a semantic manner. The proposed mapping algorithm represents a starting point to transform our proposed intelligent discovery mechanism into an applicable one. Without trying to benefit from the already available web services, any semantic discovery mechanism will face many problems to be widely used. So, the main task for this algorithm is to help the service providers to redescribe their WSs in a semantic manner. There are many advantages of using OOP-based approach to build our repository in the mapping, classification, and discovery (matchmaking) processes. These advantages includes: 1) enhancing the results of OSSE by using the structural refining stage which depends on the information collected by the repository insertion methodology about the new inserted ontologies; 2) enable us to propose a new SWSs classification algorithm that depends on a new concepts ranking algorithm. This ranking algorithm has three components (self, parent, and children) and it uses the structural information available through the proposed repository; 3) enable us to propose a new discovery (matchmaking) algorithm. This algorithm is a logic matchmaking algorithm that relies on measuring the semantic distance between the user requests and the available services. It uses a new proposed concepts tree for this measuring process. The concepts tree is created with the aid of the structural information collected by our repository. The concerning issues of these new classification and matchmaking algorithms will be discussed in our next papers after finishing their testing and evaluation processes. Experimental results show that the proposed algorithm has the potential to significantly reduce the time and efforts of the mapping process. Moreover, they show that the algorithm consideration of the standardization problem promises that it will have a positive impact on the discovery process as a whole. In the future, we will continue to implement and validate the rest of our proposed discovery mechanism components. REFERENCES [1] M. Burstein, C. Bussler, M. Zaremba, T. Finin, M.N. Huhns, M. Paolucci, A.P. Sheth, and S. illiams, A Semantic Web Services Architecture, IEEE Internet Computing, vol. 9, no. 5, pp , Sept./Oct [2] T. Berners-Lee, J. Hendler, and O. Lassila, The SemanticWeb, Scientific Am. Magazine, vol. 284, no. 5, pp , [3] J. Cardoso, Semantic Web Services: Theory, Tools, and Applications. Idea Group, Inc., [4] B. Sapkota, D. Roman, and D. Fensel, Distributed Web Service Discovery Architecture, Proc. Advanced Int l Conf. Telecomm. and Int l Conf. Internet and Web Applications and Services (AICT-ICIW 06), [5] T.A. Farrag and H.A. Ali, A Cluster-Based Semantic Web Services Discovery and Classification, Proc. ACME Second Int l Conf. Advanced Computer Theory and Eng., pp , [6] Web Services Description Language (WSDL), W3C Note, [7] Web Ontology Language for Services (OWL-S), W3C Member Submission, [8] D. Martin, M. Burstein, D. Mcdermott, S. Mcilraith, M. Paolucci, K. Sycara, D.L. Mcguinness, E. Sirin, and N. Srinivasan, Bringing Semantics to Web Services with OWL-S World Wide Web, vol. 10, no. 3, pp , Sept [9] Extensible Markup Language (XML), W3C Recommendation: [10] Simple Object Access Protocol (SOAP) Version 1.2, W3C Recommendation (second ed.) Apr [11] XML Schema, W3C Recommendation, XML/Schema, [12] Web Services Description Language version 2 (WSDL 2.0v), W3C Note: [13] B. Di Martino, Semantic Web Services Discovery Based on Structural Ontology Matching, Int l J. Web and Grid Services, vol. 5, no. 1, pp , [14] M. Klusch, B. Fries, and K. Sycara, OWLS-MX: A Hybrid Semantic Web Service Matchmaker for OWL-S Services J. Web Semantics: Science, Services and Agents on the World Wide Web, vol. 7, no. 2, Apr [15] A. Heß, E. Johnston, and N. Kushmerick, ASSAM: A Tool for Semi-Automatically Annotating Semantic Web Services, Prc. 12th Int l Conf. Web Technologies, pp , [16] A.A. Patil, S.A. Oundhakar, A.P. Shethand, and K. Verma, METEOR-S Web Services Annotation Framework, Proc. 13th Int L Conf. World Wide Web (WWW), [17] N. Oldham, C. Thomas, A. Sheth, and K. Verma, METEOR-S Web Service Annotation Framework with Machine Learning Classification, Proc. First Int l Workshop Semantic Web Services and Web Process Composition, pp , [18] M. Paolucci, N. Srinivasan, K. Sycara, and T. Nishimura, Towards a Semantic Choreography of Web Services: From WSDL to DAML-S, Proc. First Int l Conf. Web Services (ICWS 03), pp 22-26, June 2003.

13 FARRAG ET AL.: TOWARD SWSS DISCOVERY: MAPPING FROM WSDL TO OWL-S BASED ON ONTOLOGY SEARCH AND [19] OWL, [20] G. Meditskos and N. Bassiliades, Object-Oriented Similarity Measures for Semantic Web Service Matchmaking, Proc. Fifth European Conf. Web Services (ECOWS 07), [21] G. Meditskos and N. Bassiliades, Structural and Role-Oriented Web Service Discovery with Taxonomies in OWL-S IEEE Trans. Knowledge and Data Eng., vol. 22, no. 2, pp , Feb [22] W.V. Siricharoen, Ontology Modeling and Object Modeling in Software Engineering, Int l J. Software Eng. and Its Applications, vol. 3, no. 1, pp , Jan [23] P. Bartalos and M. Bielikova, An Approach to Object-Ontology Mapping, Proc. Second IFIP Central and East European Conf. Software Eng. Techniques, [24] H. Knublauch, A Semantic Web Primer for Object-Oriented Software Developers, W3C Working Group Note, [25] WordNet: An Electronic Lexical Database, C. Fellbaum ed. MIT Press, [26] Resource Description Framework (RDF), W3C Recommendation: [27] RDF Vocabulary Description Language 1.0: RDF Schema (RDFS), W3C Recommendation: [28] Database. MIT Press, OWLS-TC v3. 0, semwebcentral. org/projects/owls-tc/, [29]. Web Service Modeling Ontology (WSMO), W3C Member Submission: [30] Web Service Semantics (WSDL-S), W3C Member Submission: [31]. Semantic Annotations for WSDL and XML Schema (SAWSDL), W3C Member Submission: [32] Semantic Web Services Framework (SWSF), W3C Member Submission: [33] K. Kritikos and D. Plexousakis, OWL-Q for Semantic QoS-Based Web Service Description and Discovery, Proc. Workshop Service Matchmaking and Resource (SMRR), Tamer Ahmed Farrag received the BSc degree from the Computers and Systems Engineering Department, with general grade Excellent, Honor, 2002, and the master s degree in 2006; his thesis was entitled Scheduling Mechanism for Mobile Computing from Mansoura University. He is working toward the PhD degree at Mansoura University; his PhD thesis is entitled An Intelligent Discovery and Classification Mechanism For Semantic Web Services. He has a good knowledge of networks hardware and software. He is currently an assistant lecturer in the Department of Communications and Electronics, Misr High Institute of Engineering and Technology. Ahmed Ibrahim Saleh received the BSc degree from the Department of Computer Engineering and Systems Department in the Faculty of Engineering at Mansoura University, with general grade Excellent and the master s degree in the area of mobile agents. He has a good knowledge in networks hardware and software. He is currently a teacher at the Faculty of Engineering, Mansoura University, Egypt. His interests are programming languages, networks and system administration, and database. Hesham Arafat Ali received the BSc degree in electronics engineering, and the MSc and PhD degrees in computer engineering and control from the Faculty of Engineering, Mansoura University, in 1986, 1991, and 1997, respectively. He is a professor in computer engineering and systems and an associate professor in information systems and computer engineering. He was an assistant professor at the University of Mansoura, Faculty of Computer Science from 1997 to From January 2000 to September 2001, he was a visiting professor in the Department of Computer Science, University of Connecticut. From 2002 to 2004, he was a vice dean for student affair in the Faculty of Computer Science and Information, University of Mansoura. He received the Highly Commended Award from Emerald Literati Club 2002 for his research on network security. He is a founding member of the IEEE SMC Society Technical Committee on Enterprise Information Systems (EIS). He has many book chapters published by international press and about 150 published papers in international conferences and journals. He has served as a reviewer for many high quality journals, including the Journal of Engineering published by Mansoura University. His research interests are in the areas of network security, mobile agent, network management, search engine, pattern recognition, distributed databases, and performance analysis.. For more information on this or any other computing topic, please visit our Digital Library at