In this paper we present an ontology-based approach to support interoperability in interorganizational applications. The ontology contains knowledge, coming from several organizations, structured in di®erent layers to support its e®ective use and communication. The proposed approach has been experimented in the framework of the Italian VISPO (Virtual district Internet-based Service PlatfOrm) project.
In recent years, organizations are increasingly looking for opportunities to exploit innovative technologies that use Internet, mobile and wireless devices to enhance communication and cooperation in providing information and services. Modern inter- and intra-organizational applications speci¯cally need to support understanding of shared knowledge.
In literature, di®erent approaches address the problem of interoperability in
distributed dinamically evolving environments and propose ontologies as a
suitable solution [
In this paper, we propose an ontology-based approach to support interoperability in interorganizational applications according to a three-layer ontology ? This work has been partially supported by the Italian MIUR VISPO (Virtual district
Internet-based Service PlatfOrm [
This paper is organized as follows: Section 2 presents the considered context and provides motivations for the proposed approach; Section 3 presents the ontology design approach; Section 4 shows di®erent modalities to exploit the interorganizational knowledge represented in the domain ontology; Section 5 presents concluding remarks. 2
The availability of methodologies to construct ontologies is a critical issue in
supporting the communication and cooperation among di®erent organizations.
In fact, the explicit and shared representation of interorganizational knowledge
is a main prerequisite for exchanging information and services. According to the
experience gained in the VISPO Project [
An intra-organizational use of three-layer ontology in a Virtual District In this considered context, we address the use of ontology in the analysis of internal purchase orders, that is, a set of auditing activities to analyze and control the sources, the °ows and the volumes of the orders made from the departments of a company. In this case, the employees of each department usually formulates purchase requests according to a terminology and descriptions that are not completely normalized, but vary from department to department and often from employee to employee. The requests are directed and processed from the Purchase Department of the company that is responsible to collect the purchase requests and to send the orders to the suppliers. From the perspective of the analysis of internal purchase orders, the problem is to identify all the requests referred to the same product or to the same product category, since they can be possibly described di®erently in the various requests and purchase transactions. The main requirement, in this case, is the need for an internal catalog that provides standardized descriptions for the purchase requests. In this way, in fact, it is possible to evaluate the ordered amounts of a given product that results uniquely identi¯ed. The internal catalog can be constructed by integrating the terminology and the descriptions of products contained in the purchase requests. In the experimentation of our methodology in the context of the VISPO Project, an internal catalog has been constructed according to the three-layer ontology, presented in Section 3, by integrating the description of purchase requests extracted from the ERP of a selected company and the product descriptions obtained from two public industrial catalogs. The application of our methodology and, in particular, the integration with descriptions from industrial catalogs provide the following bene¯ts: (i) the descriptions and the terms identifying products are made homogeneous and are better standardized, (ii) the descriptions of products are enriched on the basis of the industrial catalog descriptions. The internal catalog can be exploited to formulate a new purchase request in terms of the global concepts the catalog contains (for instance, an hexagonal-head screwdriver - 4mm diameter) and, in this case, the tree-layer architecture is followed in a top down way, that is the ontology is browsed from the categorization layer down to the mediation layer until the desired product is identi¯ed. If a speci¯c product is required (for instance, an exagonal-head screwdriver of given brand and model) the architecture is used bottom-up to classify the product in the associated global concept. The Figure 1 shows schematically the process of standardization of the internal catalog. 2.2
An inter-organizational use of three-layer ontology in a Virtual District The relevant issue in developing the e-procurement service is that the di®erent purchase requests have to be aggregated into a single order from the eprocurement service to the supplier that provides the better conditions for the ordered items. It is supposed that the e-procurement service relies on di®erent suppliers, each of ones has its own catalog, so from the e-procurement perspective the problem is: (i) to provide a uni¯ed supplier catalog from which the client companies can decide their orders and (ii) to map the product description of the uni¯ed catalog onto the suppliers' ones. This scenario corresponds Analysis of internal purchase orders Service
Internal catalog
Integration and standardization Purchase requests descriptions
Public Catalog
Public
Catalog to the well known problem of catalog integration that we analyze with respect to an inter-organizational use of our ontology construction methodology. In general, to enable the integration of more catalogs that list products in overlapping domains, we need to identify correspondences among similar concepts and to construct uni¯ed representations of them that maintain references to the original catalogs. This allows one to search for a given product in the uni¯ed catalog and, after the product is found, accessing to the associated products listed in the supplier catalogs. The identi¯cation of correspondences is required at di®erent levels: { di®erent catalogs use di®erent product taxonomies as tables of contents; we need to establish mappings among the taxonomies since, in general, they use di®erent terminology and have di®erent organizations (see Fig. 2); { we need to mediate and to establish correspondences among di®erent representations of similar products adopted from di®erent catalogs and possibly to map them into standardized representations.
A uni¯ed catalog represented as three-layer ontology that satis¯es the presented requirements can be constructed as illustrated in the next sections, where in particular the products are represented as global concepts in the ontology. On the basis of this ontology, an e-procurement service can be designed to support the aggregation of purchase requests from the companies of the virtual district. In fact, if the requests are referred to products chosen in the uni¯ed catalog, the e-procurement service can classify the ordered products of di®erent purchase requests, aggregate the requests related to homogeneous products into single purchasing transactions that will regard bigger quantities, so obtaining greater discounts from suppliers. In this scenario, our thee-layer ontology can be used according to a top-down perspective, where the e-service browses the ontology Catalog 1
Hexagonal
head Catalog 2
Phillipshead Screwdriver 5mm Phillipshead 4mm 5mm 6mm 4mm 5mm 6mm 4mm 5mm 6mm Hexagonalhead
Phillipshead
Slot-head
Hexagonalhead
Slot-head
Hexagonalhead
Phillipshead
Slot-head starting from the global concepts and reformulates a purchase request made in terms of global concepts into one referring products in a supplier catalog. The application of ontology to this context is schematically shown in the Figure 3.
6mm
Supplier A Catalog A Catalog B
Supplier B
Supplier …
Catalog …
To support interoperability and communication in a virtual district scenario,
we proposed a three-layer ontology architecture which provides a uni¯ed
semantic representation of interorganizational knowledge in the considered domain by
means of ontological concepts and semantic relationships between them [
Ontological concepts. The three layers of the ontology architecture
organize interorganizational knowledge through three main kinds of concepts:
{ XClasses, that are conceptual elements of the original datasources expressed
using a common formalism, the X-Formalism, presented in [
XClasses in di®erent datasources (mediation layer); { subject categories, belonging to available standard taxonomies (categorization layer).
These di®erent kinds of concepts are speci¯ed as follows. We assume that
information in the considered datasources is expressed using XML-based or
XMLcompliant schema languages. An XClass is thus described by a name, a set
of properties or attributes (with simple or built-in data types, such as string,
NMTOKEN or integer, and some cardinality constraints) and a set of references to
other XClasses (with cardinality constraints). A global XClass is associated to a
cluster of XClasses, that are grouped on the basis of their semantic similarity [
Figure 4 shows a portion of three-layer ontology built in the VISPO project to support e-procurement services in a virtual district operating in the industrial supply market. We have considered two on-line industrial catalogs, the Beta1 and Usag2 catalogs, containing detailed descriptions of their products, and a third catalog provided by a company of the virtual district with less detailed product descriptions. From each catalog, we extracted the product descriptions, represented through XClasses, and we grouped them into clusters on the basis of their similarity in the semantic mapping layer (for example, the XClasses Toggle joint shears for sheet-steel and Shears for sheet-steel). Then, we uni¯ed descriptions belonging to the same clusters into global concepts in the mediation layer (for example, the concept Shears), ¯nding semantic relationships between them (e.g., generalization relationships between Shears and Blades). Finally, global concepts have been related to subject categories of the UNSPSC taxonomy in the categorization layer.
Intra-layer and inter-layer links. XClasses, global concepts and subject categories at di®erent levels of abstraction are related by means of inter-layer 1 \http://www.beta-tools.com/" 2 \http://www.usag.it/"
Hand Tools Forming tools
Wrenches
and drivers
and intra-layer links that can be used to browse the three-layer ontology as
explained in Section 4. In the semantic mapping layer, local XClasses belonging
to the same cluster are related each other by means of similar-to relationships,
obtained through the evaluation of their structural and name a±nity [
Global concepts in the mediation layer are organized by means of semantic relationships. We consider three kinds of relationships: (i) generalization, a concept ® generalizes another concept ¯ if the set of instances of ® includes the set of instances of ¯; (ii) disjunction, two concepts ® and ¯ are disjoint if the sets of their instances are disjoint; (iii) equivalence, two concepts ® and ¯ are equivalent if the sets of their instances coincide.
Finally, association links are maintained between global concepts and subject categories in the categorization layer. Subject categories are organized in a generalization taxonomy.
Ontological elements and intra/inter-layer links are represented in a common manner by means of the frame structure shown in Table 1, that can be easily exploited by a software agent for the ontology deployment (see Section 4), apart from the logical language used to implement the ontology. Note that not all the ¯elds of the frame structure are always mandatory for every ontological element. Source links for a global concept are obtained by means of the union of source links for local XClasses in the associated cluster; in the same manner, source links for subject categories are the union of source links for associated global concepts (in the case of leaf subject categories) or of source links for the specialized subject categories (otherwise in the taxonomy).
Property
Name of the ontological element XClass, global concept or subject category List of properties of the global concept or XClass, with associated
types and cardinality constraints; empty for subject categories Names of elements that generalize the current one (generaliza
tion relationship in the mediation and categorization layer) Names of ontological elements that are related to the current one through an equivalence relationship between global concepts in the mediation layer and a similar-to relationship between local XClasses in the semantic mapping layer Names of elements that are disjoint from the current one, in the
case of global concepts in the mediation layer Names of elements that are related to the current one by means
of an association link (inter-layer links)
Links to the datasources to which the current element is related
In [
The construction of the three-layer ontology is supported by artemis [
Domain knowledge ontology is a very useful tool to provide an access point to the interorganizational information within the virtual district and support discovery of information for business purposes. Several searching modalities can be exploited in the three-layer ontology, taking into consideration all di®erent levels of abstraction, experience of user in the considered domain and the kind of the OODDLLI3I3 XXtfotofoorOrmODmDaLalLilIs3iIms3m
XXMMLLtoto
XXfoforrmmaalilsimsm Ontology Serching
Subject Categories
Description CORBA object CORBA interaction Access to External Data
TTeerrmmininoolologgicicaall RReelalatitoionnshshipipss KKeeyywwoorrddddrriviveenn
seseaarrcchh QQuueerryyddrriviveenn seseaarrcchh
NNaammeeAAfffifninitiyty eevvaaluluaatitoionn SStrtruucctuturraall Affinity
Affinity eevvaaluluaatitoionn EExxtetennsisoionnaall
Affinity
Affinity eevvaaluluaatitoionn
Affinity evaluation GGlolobbaall AAfffifninitiyty eevvaaluluaatitoionn AARRTTEEMMISISGGUUII
QQuueerryy CCoommppoossitiitoionn
GGUUII DDoommaaininOOnntotolologgyyGGeenneerraatotorr CClulustseterriningg UUnnifiifcicaatitoionn OODDLLI3I3toto eexxtetennddeedd XXfoforrmmaalilsimsm wwrraappppeerr user: human or software agent. The di®erent searching modalities rely on a set of primitives that permit to software agents or applications to exploit ontology searching capabilities and that are listed in Table 2.
Category-driven navigation. The taxonomy of subject categories can be browsed to ¯nd the desired one and visit datasources directly connected to it; it is also possible to reach ontological concepts associated to that category and use other searching modalities. According to the portion of ontology shown in Figure 4, the user can browse the UNSPSC taxonomy from the Tool and General Machinery category to the Wrenches and drivers category and visualize all the datasources associated to the second one; the underlying tool environment supports the user executing the primitives (1) and (2) to browse the taxonomy and the primitive (7) to visualize datasources.
Concept-driven navigation. Global concepts of the mediation layer are
exploited as starting points to build queries on the global concepts in the
considered domain, propagating these queries towards each involved datasources; in
this manner, only one query is requested on the global concepts describing the
domain, instead of multiple queries submitted to all datasources; moreover, it
is not necessary to know location, terminology and content for each datasource;
results of the queries are obtained exploiting association links between global
concepts and local XClasses and are combined to obtain the query answer to
be presented to the user; we consider a query language proposed by ODMG-93
standard [
This query contains a set of global concepts, a boolean condition on the attributes of them and returns the instances of concepts involved in the query that satisfy the condition. To perform query reformulation and merging of the answers obtained from the datasources, we exploit the mapping rules de¯ned in the mediation scheme, where correspondences between global and local features are represented. In the considered example, suppose that the user retrieves the ontological concept Blades through one of the di®erent searching modalities; at this point, he searches for other ontological concepts that are related to Blades by means of semantic relationships (the system uses the primitives (4) and (7)) and chooses the concepts related to the current one by the generalization relationship (in the example, only the concept Shears); ¯nally, he requires datasources related to all visited ontological concepts by means of the primitive (7).
Keyword-driven search. A traditional keyword-driven search can be used to ¯nd desired ontological concepts (both at global and at local level of abstraction); users can specify one or more keywords and these terms are matched against the ontology, comparing them with names of ontological elements: (i) in case of names of global concepts, sources related to the matching concept and to its specializations are considered and the list of URL of these sources is presented to the user; (ii) in case of names of subject categories, sources related to the matching category and to its subcategories are considered and the list of URL of these sources is presented to the user. When more than one keyword is speci¯ed, the search is repeated for each keyword; in particular, if the keywords are concatenated by the AND operator, we consider the intersection of results obtained for single keywords, while if the keywords are concatenated by the OR operator, we consider the union of the results. As an example, the user speci¯es the keyword Screwdriver; the system visualizes all ontological elements that contain the word screwdriver or its synonyms using the primitive (3); the result of the search is the ontological concept Screwdriver and local XClasses belonging to the associated cluster; the system visualizes the corresponding datasources using the primitive (7).
Experience of the user about the considered domain and the ontology structure in°uences the choice of the searching modality: users with less experience generally prefer the category-driven or the keyword-driven search, while the concept-driven search requires more knowledge about query formulation (even if graphical interfaces are used to support this task) and about use of SQL-like query languages. An ontology-based search engine (for which we propose an architecture, described in Figure 6) provides users with an overview of categories, concepts and relevant relationships in the considered domain, supporting them in determining the information that better matches their needs. This kind of search engine can be considered to be complementary to traditional Web search engines which are based only on keyword occurences.
Navigation interface Navigation
module Ontology−based search engine
Query composition
interface Query−driven search module Search results
Keyword insertion interface Keyword−driven search module
In this paper we have proposed an ontology-based approach to support interoperability and communication in interorganizational applications, where di®erent organizations cooperate and share information to expand their market possibilities. Our approach organizes ontologies in three abstraction levels to represent the interorganizational knowledge, providing a global view on heterogeneous datasources, where the same information can be represented in di®erent ways. Several searching modalities to exploit the proposed ontological representation of knowledge are provided. A semi-automated tool environment both for ontology design and for searching support is being completed.
Future activities are devoted to the re¯nement of the representation language to express ontological concepts and semantical relationships at each level of the ontology and to the study of methods and technologies for domain ontology test, reuse and consistency checking, also in the framework of the European INTEROP network of excellence.