Discovery-Driven Ontology Evolution Silvana Castano, Alfio Ferrara, Guillermo N. Hess Dipartamento di Informatica e Comunicazione Università degli Studi di Milano Milan, Italy 20135 Email: {castano,ferrara,hess}@dico.unimi.it Telephone: (39) 0250316319 Fax: (39) 0250316229 Abstract— In this paper, we present a methodology for ontology Section III. A discussion of the activity of concept discovery, evolution, by focusing on the specific case of multimedia ontology together with an example is provided in Section IV. Section V evolution. In particular, we discuss the situation where the presents the related work in the field of ontology evolution. ontology needs to be enriched because it does not contain any concept that could be used to explain a new multimedia resource. Conclusions and future work are discussed in Section VI. The paper shows how ontology matching techniques can be used to enforce the discovery of new relevant concepts by probing II. R EQUIREMENTS FOR THE ONTOLOGY EVOLUTION external knowledge sources using both the information available in the multimedia resource and the knowledge contained in the In this paper, we address ontology evolution in a scenario current version of the ontology. where: • Multimedia resources are considered, which have been I. I NTRODUCTION already submitted to a semantic extraction process for Ontology evolution regards the capability of managing the associating them with appropriate metadata that describe modification of an ontology in a consistent way. An ontology their contents. Consequently, the ontology evolution is may change because the domain or the user needs have triggered by incoming multimedia resources with associ- changed [16] or simply because the shared conceptualization ated metadata information. • An initial domain ontology is available, providing a (i.e., the perspective) has been modified [13]. According to [11], we can define ontology evolution as the timely description of the domain of interest (in the worst case, adaptation of the ontology to changing requirements and the also a high-level, poorly detailed description). Such an consistent propagation of changes to the dependent artifacts. ontology is used also by the extraction process and will The problem of ontology evolution can be considered also as be continuously evolved so that the new ontology version a special case of the more general and well studied problem of is used as input for subsequent extraction and evolution belief change as described in [5]. In this respect, some of the activities, according to a bootstrapping process which is most important concepts of the belief change literature have iterated until a satisfactory quality of multimedia resource been revised in order to apply them to the ontology evolution classification is reached. In the paper, we will consider the context. A six-phase evolution process for ontology evolution Athletics domain and we will discuss ontology evolution has been proposed in [15] which is generally recognized as by providing some examples taken from this domain. • An open system approach is adopted, in that when back- a comprehensive reference methodology capable of handling the evolution of multiple ontologies. ground knowledge in the underlying ontology is not suf- ficient to interpret new incoming multimedia resources, In this paper, we focus on a specific case of evolution, a concept discovery activity is triggered to acquire new which is multimedia ontology evolution. In our approach, the knowledge from other external knowledge sources. This ontology is used to provide an interpretation for the elements way, we try to limit the human involvement of the on- extracted from multimedia resources, such as images, textual tology expert in the definition of new concepts, which is documents, video and audio. Our approach is defined in usually a manual and time consuming activity in ontology the framework of the BOEMIE EU Project [1], where the evolution [16]. reader can find further details about the semantic information extraction from multimedia resources as well as the whole Before introducing our evolution methodology, we introduce evolution methodology. In the paper, we focus on the use of some basic definition used throughout the paper: ontology matching techniques to support ontology enrichment • A multimedia resource is any kind of multimedia object when new knowledge is required to find an interpretation for (e.g., image, video, sound) from which information is a new resource. extracted. The paper is organized as follows. In Section II we describe • Each individual extracted element from a multimedia the requirements for ontology evolution. The methodology we resource is called a media object (e.g., a portion of an propose for multimedia ontology evolution is presented in image). • With each media object some metadata are associated, With respect to this example, the domain ontology con- which represent the semantic information that is extracted tains the aggregate concept Pole vault, while Horizontal bar, from the resource. Foam mat, Athlete and Pole are simple concepts. The ac- • A simple concept is a concept in the domain ontology tivity of classifying a new multimedia resource with respect which explains a media object in terms of metadata. to the ontology is, in this context, the activity of finding an • An aggregate concept is a concept in the domain on- aggregate concept in the ontology capable of explaining the tology which is defined as an aggregation of simple set of simple concepts detected in the multimedia resource concepts, such as, for instance, an event. Furthermore, semantic extraction. Such an activity is called interpretation an aggregate concept is an explanation of the multimedia of the multimedia resource, which can be performed by resource, that is an interpretation of the real event or exploiting non-standard reasoning techniques [9]. However, object described in the resource. not always the interpretation process succeeds in determining an explanation for the resource in the ontology. In particular, To illustrate these concepts, we consider an evolution sce- depending on the background ontology knowledge and on the nario in the Athletics domain characterized by the multimedia metadata information for the incoming resource, four cases resource shown in Fig. 1. In this case, four media objects are possible: i) one single aggregate concept explaining the (marked in the figure) are identified during semantic extraction multimedia resource semantics is found in the ontology; ii) which have associated metadata stating that: (1) is a foam more than one aggregate concept explaining the multimedia mattress, (2) is a pole, (3) is an horizontal bar, and (4) is an resource semantics are found in the ontology; iii) no aggregate athlete, respectively. Moreover, with respect to the event (i.e., concepts are found in the ontology capable of explaining aggregate concept) described by the image, we can recognize the semantics of the multimedia resource, but all the simple that all these media objects are part of a bigger picture, which concepts extracted from the resource are already defined in describes a pole vault event. the ontology; iv) no aggregate concepts are found in the on- tology capable of explaining the semantics of the multimedia resource, and one or more simple concepts are missing in the ontology too. These four scenarios produce two main classes of evolution patterns, namely population and enrichment, respectively (see Fig. 2). In the following we briefly describe these two classes of patterns. Evolution Pattern Population Pattern Single explanation of the multimedia resource Multiple explanation of the multimedia resource Enrichment Pattern Missing explanation of the Fig. 1. An example of multimedia resource multimedia resource with metadata information Missing explanation of the multimedia resource without The metadata associated with the media objects of Fig. 1 metadata information are shown in Table I. Fig. 2. The evolution patterns TABLE I M ETADATA ASSOCIATED WITH THE MEDIA OBJECTS OF THE MULTIMEDIA RESOURCE OF F IG . 1 • Population patterns: the domain ontology contains the aggregate concepts as well as all the simple concepts Foam mat f oam mat1 described by the media objects metadata extracted from Pole pole1 the multimedia resource. In this case, only ontology Horizontal bar horizontal bar1 Athlete athlete1 population has to be performed. Ontology population is the activity of introducing new instances in the ontology to describe the multimedia resource with respect to the ontology. This internal concept detection activity is optional aggregate concepts already present in the ontology. and aims at finding in the domain ontology some aggregate • Enrichment patterns: The ontology does not contain concepts which have semantic affinity with ct to be used as concepts that can be used to explain the multimedia the starting basis in the next phase. resource. In this case, the ontology has to be enriched. Concept discovery: if internal concepts have been detected in Ontology enrichment is the activity of defining new con- the previous phase, for each retrieved matching concept ci in cepts, properties, and/or semantic relations in the ontol- the ontology (i.e., ci is semantically similar to the temporary ogy. Basically, the idea behind our approach to ontology concept ct ) a probe query is created. Otherwise, the ontology enrichment is to exploit the metadata available in the expert has to define the probe query manually, by exploiting multimedia resource description in order to discover new his own domain knowledge and the information available in aggregate concepts in external knowledge sources (e.g., the definition of the temporary concept ct . A probe query is a other ontologies, web directories, lexical systems) that description of the properties, the constraints, and the semantic can be added to the domain ontology in order to explain relations that should be exhibited by the (actually missing) the new multimedia resource. This activity is referred new concept in order to consider it as a candidate for ontology as concept discovery and it is executed by exploiting enrichment. The probe query is then matched against one or ontology matching techniques. more external knowledge sources to discover some external In this paper, we focus on the enrichment patterns and, in concepts that could be useful to define the new concept to be particular, we describe the case when all the simple concepts inserted in the ontology. are already stored in the domain ontology, but the aggregate Ontology enrichment: this phase comprises the definition concept is missing, in order to show how the ontology can be and insertion of the new (aggregate) concept in the domain enriched by the concept discovery activity. ontology and also the validation of the resulting ontology. The definition of the new concept is interactively performed III. M ETHODOLOGY FOR DISCOVERY- DRIVEN ONTOLOGY by the ontology expert, by exploiting the definitions of the EVOLUTION temporary concept ct and of the external matching concepts In the evolution scenario discussed in the paper, the in- retrieved during the concept discovery phase. Once the new terpretation of a multimedia resource produces a description aggregate concept has been defined, the ontology expert inserts of the simple concept instances that are detected into the it in a proper position in the ontology taxonomy. The task of resource (e.g., the objects that appear into an image) but no finding the most suitable placement of the new concept in aggregate concept (e.g., events) in the ontology can be found the ontology taxonomy is performed by combining ontology that explain the resource itself. In this section, we describe matching techniques and ontology reasoning, as described a methodology for ontology evolution in this scenario and in [2]. Finally, the evolved ontology is validated by means the role of ontology matching techniques in supporting the of standard reasoning techniques [6]. identified evolution phases. B. Ontology matching for concept discovery A. The evolution phases As described in the previous section, ontology matching The goal of ontology evolution is to augment the back- techniques play an essential role in the ontology evolution ground knowledge in the existing domain ontology to better process. In temporary concept definition, the temporary con- classify extracted object descriptions. The ontology is enriched cept can be compared against the existing ontology to find by adding concepts, properties, and/or semantic relations or by internal matching concepts that can help the probe query modifying existing ones, in order to produce a new ontology formulation activity. In concept discovery, a probe query is version capable of providing interpretation of the new multi- compared against external knowledge sources with the goal media resource. of finding potentially relevant knowledge. To perform these In order to detect the ontology modifications required for concept comparisons, we rely on our semantic matchmaker explaining the new resources, we exploit the description of the H-M ATCH developed in the framework of the H ELIOS project media objects extracted from the resources. This means that, for matching independent peer ontologies [3]. H-M ATCH takes different from the usual evolution scenarios, in our case we a target concept description c and an ontology O as input know from the interpretation phase that an aggregate concept and returns the concepts in O which match c, namely the is missing in the domain ontology, but we do not know exactly concepts with the same or the closest intended meaning of c. which one. We have the instance of the missing concept and In H-M ATCH we perform concept matching through affinity have to create an aggregate concept which explains it. metrics by determining a measure of semantic affinity in The evolution process is articulated in three phases, which the range [0, 1]. A threshold-based mechanism is enforced are explained in the following. to set the minimum level of semantic affinity required to Temporary concept definition: the incoming instance and its consider two concepts as matching concepts. Given two con- associated metadata information is used to define a temporary cepts c and c0 , H-M ATCH calculates a semantic affinity value concept ct , which is matched against the domain ontology SA(c, c0 ) as the linear combination of a linguistic affinity to detect semantically related concepts already present in the value LA(c, c0 ) and a contextual affinity value CA(c, c0 ). The linguistic affinity function of H-M ATCH provides a measure of H-M ATCH is executed by exploiting only the contex- similarity between two ontology concepts c and c0 computed tual affinity matching techniques. In fact, contextual on the basis of their linguistic features (i.e., concept names). matching considers only the contextual affinity between For the linguistic affinity evaluation, H-M ATCH relies on the concepts to be compared without considering their a thesaurus of terms and terminological relationships auto- names. The goal of this activity is to exploit the concepts matically extracted from the WordNet lexical system [12]. already inside the ontology which are similar to ct , and The contextual affinity function of H-M ATCH provides a which are featured by a name, to trigger the concept measure of similarity by taking into account the contextual discovery phase. In other terms, we know that the features of the ontology concepts c and c0 . The context of a missing concept is similar to something that is already concept can include properties, semantic relations with other known in the ontology, and we exploit this similarity concepts, and property values. The context can be differently between the ontology (what we know) and the external composed to consider different levels of semantic complexity, knowledge sources (what is known in other sources) in and four matching models, namely, surface, shallow, deep, and order to learn new concepts to be added to the ontology. intensive, are defined to this end. In the surface matching, only 3) For each concept ci whose affinity with ct is higher the linguistic affinity between the concept names of c and c0 than a threshold t, we build a probe query qi . Each is considered to determine concept similarity. In the shallow, probe query qi is matched against external knowledge deep, and intensive matching, also contextual affinity is taken sources using H-M ATCH. The result is a set of external into account to determine concept similarity. In particular, candidate concepts returned by the matching process the shallow matching computes the contextual affinity by whose semantic affinity is higher than or equal to a given considering the context of c and c0 as composed only by their matching threshold. Retrieved concepts are then made properties. Deep and intensive matching extend the depth of available to the ontology expert, who can exploit them concept context for the contextual affinity evaluation of c and to define the new concept to be added to the ontology. c0 , by considering also semantic relations with other concepts (deep matching model) as well as property values (intensive function conceptDiscovery(Extracted_Metadata){ matching model), respectively. The comprehensive semantic Vector probeQueries = null; affinity SA(c, c0 ) is evaluated as the weighted sum of the Vector candidateConcepts = null; TemporaryConcept ct = createConcept(Extracted_Metadata); Linguistic Affinity value and the Contextual Affinity value, foreach Concept c in Ontology{ that is: if(hmatch.contextualMatch(ct,c) >= threshold){ probeQueries.add(c); SA(c, c0 ) = WLA · LA(c, c0 ) + (1 − WLA ) · CA(c, c0 ) (1) } foreach ProbeQuery q in probeQueries{ foreach Concept e in ExternalKnowledgeSource{ where WLA is a weight expressing the relevance to be given if(hmatch.match(q,e) >= threshold){ for the linguistic affinity in the semantic affinity evaluation candidateConcepts.add(e); } process. A detailed description of H-M ATCH and related } matching models is provided in [3]. } } return candidateConcepts; IV. C ONCEPT DISCOVERY AND ONTOLOGY ENRICHMENT } In this section, we describe in more detail how concept dis- Fig. 3. Concept discovery algorithm covery and ontology enrichment are performed, with respect to the jumping example of Fig. 1. In order to discuss an example of concept discovery, con- A. Concept discovery sider the multimedia resource shown in Fig. 1 and the media objects marked in the figure. Then suppose that the domain on- The concept discovery activity is articulated in the following tology contains information about jumping events, as defined steps: in Fig. 4. The concepts graphically represented by a white 1) Given a multimedia resource r and the set M of media box (i.e., Sport event, Athletic event, Jumping event and objects that have been extracted from r, we build a High jump) represent aggregate concepts, while gray boxes temporary target concept ct . The context of ct (i.e., (e.g., Foam mat, Pole, Horizontal bar, Athlete) represent the set of properties, restrictions, and semantic relations simple concepts. Composition dependencies among simple and featuring ct ) is composed by exploiting the metadata aggregate concepts are represented by means of dashed arrows. associated with M , that is the set of simple concepts S In the right side of the figure, the corresponding Description explaining objects in M , that is, Logics specification is reported. S = {ci ∈ O | ∀mi ∈ M, mi : ci } The first step of concept discovery is the creation of the temporary concept ct , starting from the incoming resource where O denotes the ontology. information, as shown in Fig. 5. 2) We use H-M ATCH to match ct against the existing The definition of ct is based on the analysis of the metadata domain ontology. Since ct is not featured by a name, associated with the multimedia resource, that are shown in Athlete ≡ P erson u ∃hasP rof ession.Sport F oam mat v SportEquipment P ole v SportEquipment Javelin v SportEquipment Horizontal bar v SportEquipment Jumping event v Event High Jump v Jumping event u ∃hasP art.Jumper u ∃hasP art.Horizontal bar u ∃hasP art.F oam mat Fig. 4. The domain ontology before evolution ct v ∃hasP art.Jumper u B. Ontology enrichment ∃hasP art.Horizontal bar u Based on the matching concepts definitions retrieved during ∃hasP art.F oam mat u the concept discovery phase, on the incoming temporary ∃hasP art.P ole concept ct and on the domain ontology, the ontology expert Fig. 5. The temporary concept ct definition from the metadata of Table I begins the concept definition activity for ontology enrichment. A new aggregate concept is defined using the simple concepts available in the domain ontology. If some simple concept is Table I. In particular, since the media object retrieved in needed which is not yet described in the ontology, it is also the resource are considered as components of an unknown defined. In the example, the ontology expert chooses to add the concepts that should be added in the ontology, we define on Pole Vault concept, on the basis of the information provided ct a hasP art restriction for each simple concept of Table I. by the two web sources and on the fact that Pole Vault is The temporary concept ct describes the features (in terms of retrieved in both the external sources. Concept definition is a restrictions) of the required target concept. However, we do not manual activity, in that the ontology expert must specify all the have neither a name for ct nor any semantic relation among it properties and restrictions of the new concept. In our approach, and other concepts in the ontology. Because of this, if we the amount of manual activity is reduced in that the ontology match ct against the external knowledge source, we could expert works directly on the concepts specifications retrieved not find relevant candidate concepts. In order to address this during the discovery activity, by properly integrating/merging problem, we enrich the definition of ct by exploiting the name those considered more relevant. Once the new concept is and the context of other concepts, if any, in the ontology that defined, the ontology expert inserts it into the ontology, which are semantically similar to ct . Since a name is not available means that the appropriate location in the ontology hierarchy for ct , only the contextual matching is performed. In the case has to be found. To this end, the ontology expert may use of the example, H-M ATCH evaluates the following semantic H-M ATCH to locate a possible insertion point for the new affinity values in the ontology of Fig. 4: SA(Athlete, ct ) = concept and a reasoner to validate the new resulting version 0.45, and SA(High Jump, ct ) = 0.97. Using the H-M ATCH of the ontology [2]. Fig. 7 show the definition of the new default threshold of 0.5, only the High jump concept is aggregate concept Pole vault from our example. actually selected for probe query composition. Having the High jump concept retrieved, we now create one probe query V. R ELATED W ORK for High jump (i.e., q1 ) to be compared against external Ontology evolution research work has mainly focused on knowledge sources. In our example, we use the Google and the problem of evaluating the impact of requirement changes the Yahoo sports directories as external knowledge sources. on the ontology contents [8]. In [10], the authors present a The probe query q1 contains a description of High jump, framework for ontology evolution and change management which is composed by the name of the concept and by its based on an ontology of change operations with the aim of context, i.e., by the restrictions defined on it and by its super- providing a formal description of the ontology modifications classes and subclasses. The result of matching q1 against the required to perform a given evolution task. The ontology of Google and Yahoo external sources is shown if Fig. 6, where change operations is defined for the OWL knowledge model matching concepts are shown together with corresponding and contains basic change operations and complex change semantic affinity values. operations. A basic operation describes the procedure of mod- All the 11 candidates concepts shown in Fig. 6 are returned ifying only one specific feature of the OWL knowledge model to the ontology expert. High jump is discarded, since it is (e.g., type and cardinality restriction change), while a complex already present in the domain ontology. operation describes an articulated change procedure and is Fig. 6. Sample of probe query evaluation Athlete ≡ P erson u ∃hasP rof ession.Sport F oam mat v SportEquipment P ole v SportEquipment Javelin v SportEquipment Horizontal bar v SportEquipment Jumping event v Event High Jump v Jumping event u ∃hasP art.Jumper u ∃hasP art.Horizontal bar u ∃hasP art.F oam mat P ole vault v Jumping event u ∃hasP art.Jumper u ∃hasP art.Horizontal bar u ∃hasP art.F oam mat ∃hasP art.P ole Fig. 7. The domain ontology after evolution composed of multiple basic operations. With respect to these of semiautomatic/automatic ontology evolution strategies. An classification of changes, the requirements that an ontology overview of some proposed approaches in this direction is management tool should address for ontology evolution are presented in [4], even if limited concrete results have appeared discussed in [17]. In particular, the authors emphasize that such in the literature. In most recent work, formal and logic-based a tool should provide a set of evolution operations according approaches to ontology evolution are also being proposed. to the supported ontology models (functional requirement) In [7], the authors provide a formal model for handling the and that changes should be discovered semi-automatically by semantics of change phase embedded in the evolution process analyzing user behavior (refinement requirement). During the of an OWL ontology. The proposed formalization allows to evolution process, the tool has to reflect the user preferences define and to preserve arbitrary consistency conditions (i.e., (user supervision requirement) by providing advanced facili- structural, logical, and user-defined). ties, such as change-visualization and inconsistency detection A six-phase evolution methodology has been implemented (transparency and usability requirements). Moreover, history within the KAON [14] infrastructure for business-oriented of changes needs to be supported to eventually undo any ontology management. The ontology evolution process starts change applied to the ontology (auditing and reversibility with the capturing phase, that identifies the ontology modifi- requirements). The recent success of distributed and dynamic cations to apply either from the explicit business requirements infrastructures for knowledge sharing has raised the need or from the results of a change discovery activity. In the representation phase, the identified changes are described in a R EFERENCES suitable format according to the specification language of the [1] BOEMIE Project, “Bootstrapping ontology evolution with multimedia ontology to modify (e.g., OWL). The effects of the changes information extraction,” IST-2004-2.4.7 Semantic-based Knowledge and are evaluated in the semantics of change phase, where the Content Systems Proposal, 2004. [2] S. Castano, A. Ferrara, and S. Montanelli, “Evolving open and inde- ontology consistency check is also performed. Due to the pendent ontologies,” Journal of Metadata, Semantics and Ontologies, fact that an ontology can reuse or extend other ontologies 2006. (e.g., through inclusion or mapping), the propagation phase [3] ——, “Matching ontologies in open networked systems: Techniques and applications,” Journal on Data Semantics (JoDS), vol. V, 2006. ensures that any ontology change is propagated to the possible [4] Y. Ding and S. Foo, “Ontology research and development. part 2 - dependent artifacts in order to preserve the overall consistency. a review of ontology mapping and evolving,” Journal of Information The subsequent implementation phase has the role to log Science, vol. 28, no. 5, pp. 375–388, 2002. [5] G. Flouris, D. Plexousakis, and G. Antoniou, “Evolving ontology all the performed changes in order to support the recovery evolution.” in SOFSEM, ser. Lecture Notes in Computer Science, J. Wie- facilities that in the final validation phase are provided to dermann, G. Tel, J. Pokorný, M. Bieliková, and J. Stuller, Eds., vol. reverse an ontology change in case that an undesired effect 3831. Springer, 2006, pp. 14–29. [6] V. Haarslev and R. Möller, “Description of the RACER system and its occurs. applications.” CEUR Electronic Workshop Proceedings, http://ceur- With respect to the state of the art on ontology evolution, ws.org/Vol-49/, 2001, pp. 132–141. [7] P. Haase and L. Stojanovic, “Consistent evolution of owl ontologies.” in the original contribution of the presented approach is an ESWC, ser. Lecture Notes in Computer Science, A. Gómez-Pérez and evolution methodology tailored to provide a semi-automated J. Euzenat, Eds., vol. 3532. Springer, 2005, pp. 182–197. support to the new concept detection and definition activities, [8] P. Haase and Y. Sure, “State-of-the-art on ontology evolution,” Institute AIFB, University of Karlsruhe, SEKT informal which are conventionally the most human-intensive activities. deliverable 3.1.1.b, 2004. [Online]. Available: http://www.aifb.uni- In particular, we address the problem of semi-automatically karlsruhe.de/WBS/ysu/publications/SEKT-D3.1.1.b.pdf discovering the new concepts required in the ontology to [9] J. R. Hobbs, M. Stickel, D. Appelt, and P. Martin, “Interpretation as abduction,” Artificial Intelligence Journal, vol. 63, pp. 69–142, 1993. describe new multimedia resources by exploiting ontology [10] M. Klein and N. Noy, “A component-based framework matching techniques. for ontology evolution,” 2003. [Online]. Available: cite- seer.ist.psu.edu/article/klein03componentbased.html [11] A. Maedche, B. Motik, L. Stojanovic, R. Studer, and R. Volz, “An VI. C ONCLUSION infrastructure for searching, reusing and evolving distributed ontologies,” in WWW. ACM, 2003, pp. 439–448. In this paper, we have presented a methodology for sup- [12] G. A. Miller, “WordNet: A Lexical Database for English,” Communica- porting the ontology enrichment activity in the context of tions of the ACM (CACM), vol. 38, no. 11, pp. 39–41, 1995. [13] N. F. Noy and M. Klein, “Ontology evolution: Not the same as schema multimedia ontology evolution. This problem and its peculiar evolution,” Knowl. Inf. Syst., vol. 6, no. 4, pp. 428–440, 2004. requirements, is one of the main research issues addressed [14] D. Oberle, R. Volz, B. Motik, and S. Staab, “An extensible ontology in the BOEMIE [1] project, where the discovery approach software environment,” in Handbook on Ontologies, ser. International Handbooks on Information Systems, S. Staab and R. Studer, Eds. presented in the paper is studied. The original contribution Springer, 2004, ch. III, pp. 311–333. of the work is focused on the role of ontology matching [15] L. Stojanovic, A. Madche, B. Motik, and N. Stojanovic, “User-driven techniques with respect to the ontology evolution scenario. ontology evolution management,” in European Conf. Knowledge Eng. and Management (EKAW 2002). Springer-Verlag, 2002, p. 285300. [On- We have shown how ontology matching can support the line]. Available: http://wwwred.fzi.de/wim/publikationen.php?id=804 ontology expert in retrieving new concepts to be added in the [16] L. Stojanovic, A. Maedche, N. Stojanovic, and R. Studer, “Ontology ontology. In the concept discovery activity, the use of matching evolution as reconfiguration-design problem solving.” in K-CAP, J. H. Gennari, B. W. Porter, and Y. Gil, Eds. ACM, 2003, pp. 162–171. techniques reduces the manual effort required to the ontology [17] L. Stojanovic and B. Motik, “Ontology evolution within ontology expert, by suggesting a set of alternatives (candidate concepts) editors,” in OntoWeb-SIG3 Workshop at the 13th International for the definition of the new concept. Our future work will Conference on Knowledge Engineering and Knowledge Management EKAW 2002; Siguenza (Spain), 30th September 2002, 2002. [Online]. be devoted to: i) implement and test a software tool for Available: http://km.aifb.uni-karlsruhe.de/eon2002/program.html supporting the whole discovery process; ii) study the problem of defining a new concept out of the specifications of retrieved matching concepts, by exploiting a combination of matching and reasoning techniques; iii) to investigate the problem of storing and maintaining the mappings retrieved among the ontology concept and the external knowledge sources, in order to reuse the results of an activity of concept discovery in subsequent stages of ontology enrichment. ACKNOWLEDGMENT This work is founded by the BOEMIE Project, FP6-027538 - 6th EU Framework Programme. The authors would like to thank the partners of the BOEMIE project for providing the pictures for the examples.