Applied Artificial Intelligence, 19:845–859 Copyright # 2005 Taylor & Francis Inc. ISSN: 0883-9514 print/1087-6545 online DOI: 10.1080/08839510500234222

COMPETENCE ONTOLOGY FOR DOMAIN KNOWLEDGE DISSEMINATION AND RETRIEVAL

Bernard Lefebvre, Gilles Gauthier, Serge Tadie´, Tran Huu Duc, and Hicham Achaba & Laboratoire GDAC, De´partement d’informatique, Universite´ du Que´bec a` Montre´al, Montre´al, Que´bec

& In this paper, we present a competence ontology for domain knowledge dissemination and retrieval services, which has been used in the MDKT project (Management and Dissemination of Knowledge in Telecommunication). The main objective of this project is to set up a computerized knowledge management system related to a specific domain in order to develop the human resources expertise for the needs of the enterprise. In the case of this project, the knowledge is about wireless networking and is expressed in digital documents. Among all the ontologies that implement the knowledge needed by the system, the competence ontology plays a key role. The competence ontology defines at a meta-level the concept of competence and its relationships with other concepts such as document or user. Its instantiation is used to characterize a user model and a document model. This knowledge organization makes it possible to infer which document, or more generally which domain knowledge information, is suitable for a given person or to whom specific domain knowledge information should be disseminated.

One of the aspects of the Grid is its ability to make it possible to share services on the Internet through software agents. These services have to be specified in order to be used by clients of various kinds. This specification is formally given in a standardized language called WSDL (Web Services Description Language). The published formal description can be used by a service requester to identify a suitable service provider. It makes it also possible for both partners to exchange messages allowing the requested service to be achieved by the provider according to the needs of the requester. This work is being carried out under the MDKT project (Management and Dissemination of Knowledge in Telecommunication) by researchers from UQAM (University of Quebec at Montreal) and University of Montreal including Bernard Lefebvre, Jean-Guy Meunier, Gilles Gauthier, Omar Cherkaoui, Olivier Gerbe´, and some M.Sc. and Ph.D. students. We would like to thank LUB (Laboratoire Universitaire de Bell) and NSERC (Natural Sciences and Engineering Research Council of Canada) for their financial support. Address correspondence to Bernard Lefebvre, Laboratoire GDAC, De´partement d’informatique, Universite´ du Que´bec a` Montre´al, Case Postale 8888, Succursale Centre-Ville, Montre´al, Que´bec H3C 3P8, Canada. E-mail: [email protected]

846

B. Lefebvre et al.

The knowledge of the environment in which a service is going to be rendered is also a feature that can help to configure the service itself in order to suit the particularities of a requester. In order to be usable and sharable, the knowledge related to the context of use also has to be modeled by a meta level, which can be implemented by the means of formal and common ontologies. This need for sharable semantic knowledge in the context of Grid services is analogous to the one that exists for the Web in order to focus the research of documentary information. It is then appropriate to consider and apply for the Grid the tools and languages that have been developed for the Semantic Web. The object of this paper is to present such semantically dependant services in the case of a system whose main task is the development of human resources expertises. More precisely, this system integrates documentary dissemination and retrieval services and is part of the MDKT project (Management and Dissemination of Knowledge in Telecommunication). The following sections present some methodological aspects that are related to the ongoing standardization for a formalism and a language to express domain knowledge in the context of the Semantic Web. The architecture of the system built for the MDKT project is then detailed, followed by a description of the ontologies that implement the knowledge part of the system. The emphasis is given on the competence, document, and user ontologies that are used by the dissemination and retrieval services in order to focus their response. METHODOLOGICAL ASPECTS The realization of the MDKT system relies on works related to the use of ontologies for enterprises (Barry 1998; Uschold et al. 1998) and domains of knowledge. Many of those (Abecker et al. 2000; Fox et al. 1996) refer to the work of Abecker et al. (1998) which, within the framework of the project ‘‘KnowMore,’’ has developed an architecture of organizational memory, meeting the needs for interrogation in the context of a task. While the KnowMore project mainly deals with domain knowledge to characterize documents, the MDKT project adds another dimension, the competence, which is also used to characterize users. The MDKT project shares with the Ontologging project (Razmerita et al. 2003) the objective of integrating agents into the system to contribute to a better personalization of knowledge delivery and to provide adapted content. Another original aspect of the MDKT project is its commitment to respect and use the standards proposed for the Semantic Web in order to benefit from the developments and research made in this framework. As defined by Tim Berners-Lee and his collaborators, the ‘‘Semantic Web is not a separate Web but an extension of the current one, in which

Ontology for Knowledge Retrieval

847

information is given well-defined meaning, better enabling computers and people to work in cooperation . . . For the Semantic Web to function, computers must have access to structured collections of information and sets of inference rules that they can use to conduct automated reasoning’’ (Berners-Lee et al. 2001). In order to achieve adequate inferences about knowledge, computer programs must have, for example, a way to discover the common meanings for the terms designing a single concept. A solution to this problem is provided by another basic component of the Semantic Web, collections of information called ontologies. ‘‘In philosophy, an ontology is a theory about the nature of existence, of what types of things exist; ontology as a discipline studies such theories. Artificial-intelligence and Web researchers have co-opted the term for their own jargon, and for them an ontology is a document or file that formally defines the relations among terms’’ (Berners-Lee et al. 2001). In order to be used in the context of a Grid, ontologies must be published and shared and become ‘‘common ontologies.’’ The research in the framework of the Semantic Web have several objectives: . Define languages based on XML in order to represent a knowledge that can be processed by computer. . Build ontologies to describe domains of knowledge. . Use ontologies to produce semantic annotation for documents on the Web. . Define browsers and search engines able to take into account the semantic annotations. The MDKT system realizes and integrates the aspects mentioned. Compared to the extent of the Semantic Web, the scope is however considerably reduced. One other main difference is that the system context of use can be precisely defined as well as the users’ competences and skills. These particularities are heavily used by the system in order to focus the knowledge access tools. THE ARCHITECTURE OF THE SYSTEM The architecture is organized in three blocks (Figure 1): the editing and maintenance tools, the operating services, and the knowledge model. The knowledge model in the MDKT architecture is organized in three different layers by the mean of ontologies connected to a base of document resources. The first layer is called the meta structure layer and it is based on specifications proposed by the W3C for the OWL or the DAML þ OIL languages. The second one is the structure layer, which describes the structure of the knowledge needed by the tools of the MDKT system. The third one,

848

B. Lefebvre et al.

FIGURE 1 Architecture of the MDKT system.

called the data layer, is the instantiation of the second layer. The resources part contains electronic documents or Web references to documents. In the current implementation, tools, services and most parts of the knowledge model are on a single server. In the context of the Grid they could be spread on several providers. One of these could host the operating services and the others could host the two upper layers of knowledge. The data layer is client dependent and should be instantiated on the requester side. The editing and annotation tools can also be Grid provided services, but the help system for the acquisition of new concepts is too domain dependent to be in the same situation. Meta Structure Layer This layer describes the structure of an ontology according to the W3C specifications and standards. It is the ontology meta model. It explains how classes (concepts) and properties (relations) are related and expresses all the relationships between concepts in a domain. This model is built around three main concepts: class axioms, property axioms, and individual axioms, and is defined by standard documents published by the W3 Consortium. These documents are imported as name spaces by the other layers. In addition to these standard documents, the meta structure layer also defines the notion of ‘‘concept,’’ which generalizes all the entities of the domain knowledge. The Structure Layer The structure layer contains ontologies implementing the model of the knowledge used to perform the dissemination and the retrieval of domain

Ontology for Knowledge Retrieval

849

knowledge. This layer is built using the meta structure layer and more precisely the class and the property axioms. It contains all the classes1 shown in Figure 2. These classes are organized in six different ontologies: PAT, Knowledge Domain, Document, Competence, Employee, and Company. . PAT: This ontology contains the description of all the working processes that exist in a company. The abstract class PAT is specialized into process, activity, and task. This ontology also characterizes all the relations between these items. . Domain: This ontology describes general elements and properties concerning the organization of knowledge on a domain. All these items are instances and also subclasses of the class ‘‘concept,’’ which represents all the concepts that can exist in all domains. One of the properties of ‘‘concept,’’ which is inherited by all the items of the knowledge domain, is to be quoted in a document resource. Apart from the notion of concept, the content of this ontology can be imported from an existing ontology about a domain. . Document: This ontology defines the general nature of document resources. The abstract class DocumentResource is specialized into two concrete classes: document and document part. The other definitions about the nature of documents are imported from the DublinCore namespace, which is the most used and known ontology about documents.

FIGURE 2 The knowledge model of MDKT.

850

B. Lefebvre et al.

. Competence: This ontology is the center of the knowledge model and is described in greater detail later. It makes it possible to link documents to users, processes to users, and concepts to users. . Employee: This ontology contains the concepts allowing to define the profile of any user of the system that is an employee of a company. It helps in personalizing the exchange between the system and a particular user. . Company: This ontology describes a company in particular in terms of roles. A role can be played by an employee and needs some competences. Data Layer It represents the third layer of the architecture. It contains assertions about the previous ontologies, describing a real company with employees to whom some specific documentation can be delivered to help them achieve their tasks. All the instances of the class ‘‘concept’’ are also at this layer. In the case of the MDKT project, the domain is wireless networking so that instances of ‘‘concept’’ like access-point or antenna are in this layer but are also considered as classes in a hierarchy of classes in the upper layer. For example, the former ones are both subclasses of the class ‘‘device.’’ All these classes are also subclasses of the main class ‘‘concept,’’ so that they inherit the property being quoted in a document. A particular brand of antenna, which is an instance of the class ‘‘antenna,’’ can thus be quoted in a document part as well as the class itself. Document Resources The block document resources contains the entire documentation available by the mean of the dissemination or retrieval tools. Each of these documents is designated by a URI. The resource can be in one of the many formats usable on the Internet: HTML, PDF, word, video, etc. THE COMPETENCE ONTOLOGY A competence is a know-how, a knowledge or a behavior that can directly or indirectly be measured, calculated, acquired, indicated, or examined. This behavior makes it possible for a person to carry out a task in accordance with the requirements of a situation of work. Competence exists in relation to other concepts describing the capability, the skill, and the expertise of a person. These elements and their relations, in the case of the MDKT project, constitute the model of competences that is illustrated in the example given next.

851

Ontology for Knowledge Retrieval

The relation between employees and competences in the MDKT project is not only directly specified, it exists also indirectly through the roles carried out by the employees. This relation represents the requirements an employee must fulfill when it takes part in a project or an activity of the company. There are also relations between competences themselves. Four types of relations are usually distinguished (of which only the first three are used in the MDKT project): analogy, generalization, aggregation, and deviation (Nkambou and Lefebvre 1996): . Relation of analogy: Competences can be similar from the point of view of their functionality, their result, or their definition. . Relation of generalization: Two competences bound by this type of relation, share the attributes of the most general one but the most specific one has additional attributes. . Relation of aggregation: This relation establishes the fact that a competence is a component of another. . Relation of deviation: This relation exists if a competence is a deformation of another. There are several taxonomies of competences depending on the points of view and the criteria used. Competences, in the case of the MDKT TABLE 1 The Skills Taxonomy According to Paquette Skills taxonomy layers 1 Receive

Reproduce

2 1. Pay attention 2. Integrate 3. Instantiate=Specify

4. Transpose=Translate 5. Apply Create

6. Analyze

7. Repair 8. Synthesize

Self-manage

9. Evaluate 10. Self-manage

3

2.1 2.2 3.1 3.2 3.3

Identify Memorize Illustrate Discriminate Clarify

5.1 5.2 6.1 6.2 6.3 6.4

Use Simulate Deduce Classify Predict Diagnose

8.1 Induce 8.2 Plan 8.3 Model=Construct 10.1 Influence 10.2 Self-control

852

B. Lefebvre et al.

project, are classified in two groups. They can be transverse or specific. The specific ones are directly related to the realization of a task while the transverse ones refer to competences that can be needed in the whole process of work and are often related to domain knowledge. A competence is associated to a specific level in order to be evaluated ‘‘quantitatively’’ and ‘‘qualitatively.’’ Level 1 means beginner while level 5 means expert. The combination of a competence and a level is called an expertise. Moreover, competences can be qualified by skills, which can present various levels of complexity (Paquette 2002) organized in a taxonomy (Table 1).

An Example of Assertions about Competences As previously said, an expertise level is a measurement that allows to quantify a given competence. The association between an expertise level and a qualified competence defines an expertise. In Example 1, the expertise ‘‘CharacterizeAssembly_1’’ associates the competence ‘‘dictComp_T002’’ to the level ‘‘Level2.’’ This means that the person who has the expertise ‘‘CharacterizeAssembly_1’’ is able to do ‘‘characterize assembly tools and techniques’’ at the junior level. This competence is formally qualified by the skill ‘‘skill_classify,’’ which is at the bottom of the skill taxonomy. It has at the parent level the skill ‘‘skill_analyze’’ which itself is generalized by the skill ‘‘skill_create’’ (the whole taxonomy of skills is presented in Table 1) (Paquette 2002). In Example 2, user ‘‘ROBL1’’ has expertises such as ‘‘InstallNetwork_3,’’ ‘‘InstallInterfaceCard_4,’’ or ‘‘CharacterizeAssembly_2.’’ He has the role ‘‘KO995428659670’’ in an unspecified project of the company. The expertise ‘‘InstallNetwork_3’’ is at level 3, which means ‘‘Intermediate’’ and is related to the competence ‘‘dictComp_S001: be able to install a network.’’ In the taxonomy defined in the competence ontology, this general competence is decomposed in several more specific ones such as ‘‘dictComp_S002: be able to install interface cards’’ or ‘‘dictComp_S003: be able to install software.’’ It will be inferred by the system that user ‘‘ROBL1’’ has at least these more specific expertises at the same level than the more general one. This inferred knowledge can be superseded by an explicit assertion. In Example 2, for the competence ‘‘dictComp_S002,’’ the user ‘‘ROBL1’’ has in fact a better expertise, which is ‘‘InstallInterfaceCard_4.’’ This one is associated with ‘‘Level4,’’ which corresponds to ‘‘Specialist.’’ The role ‘‘KO995428659670’’ is described in Example 3. It requires itself specific expertises: such as ‘‘InstallNetwork_1,’’ ‘‘InstallInterfaceCard_1,’’ or ‘‘CharacterizeAssembly_1.’’ It is inferred that every user having this role in the company has at least these expertises.

Ontology for Knowledge Retrieval

EXAMPLE 1 Competence assertions in file ‘‘competenceAssertions.daml.’’

EXAMPLE 2 User assertions in file ‘‘employeeAssertions.daml.’’

853

854

B. Lefebvre et al.

EXAMPLE 3 Role assertions in file ‘‘companyAssertions.daml.’’

EXAMPLE 4 Document assertions in file "documentAssertions.daml."

For a given document, in order to be able to understand it, a person needs expertises (necessary expertise). The document can bring, after being read, new expertises (acquired expertise). In Example 4, the document ‘‘Doc13’’ requires expertises such as ‘‘InstallNetwork_2’’ or ‘‘ProposeTechnology_4.’’ It helps to acquire ‘‘InstallNetwork_5’’ and ‘‘ProposeTechnology_4.’’ The filtering mechanism is used for documents and users according to the semantic relations between these concepts by ensuring that the first one is suitable for the second one. The notion of zone of proximal development presented in the next subsection is the key feature for the implementation of this mechanism.

Ontology for Knowledge Retrieval

855

The Zone of Proximal Development The zone of proximal development, according to Vygotsky (Vygotskii and Kozulin 1986; Vygotskii and Cole 1978), is ‘‘the distance between the current level of development of a person when it solves the problem all alone (without any guided instruction) and the level of potential development which is that that this person has when it solves the same problem with the assistance of an adult or when it collaborates with another more advanced person (with external assistances).’’ The concept of zone of proximal development enables to determine, with the collaboration of the competence model, the capacity of training of a person in a specific field in order to decide if such a document, which requires some expertise to be understood and which helps to acquire some expertise, is adapted or not for such a person. The distance between the zone of proximal development and the current development is called the depth of the zone. It can vary according to the fields. It is not the same for each person either. In the MDKT project, this depth is defined by the difference between the acquired and the required expertise. It is determine for each document but it does not vary from a user to another.

Semantic Reasoning The ontology languages (RDF, DAML þ OIL) have theoretical bases that are description logics. In fact, DAML þ OIL is equivalent to the SHIQ-DL (Horrocks 2002; Fensel et al. 2001). A semantic reasoning is made thanks to the implicit or explicit semantic relations between the entities or the classes represented in an ontology. It is based on the subsumption inference process. For example, a person wanting to buy roses, in a semantic research, can be connected to a florist thanks to a semantic link between the class flower (since the rose is one of its specializations) and the florist who sells them. In addition to subsumption, heuristics can be added in a semantic reasoning in order to have more ‘‘intelligent’’ and ‘‘natural’’ behaviors. For example, the set of persons having expertise ‘‘InstallNetwork_3’’ is not only composed by persons having exactly this feature but also by persons having the expertises ‘‘InstallNetwork_4’’ and ‘‘InstallNetwork_5,’’ which are at a higher level. Other reasoning heuristics, which are implemented in the system, concern the aggregation relation (‘‘isDecomposeIn’’) and the analogy (‘‘sameAs’’) between competences or skills (Figure 2).

856

B. Lefebvre et al.

REALIZATION OF THE ACTIVE DISSEMINATION AND RETRIEVAL SERVICES The active dissemination must, being given a document, be able to filter and provide to the administrator a list of users who are candidates to receive the document. It must make it possible to the expert to exclude people who are nonrelevant and finally it must be able to send a message to the selected users. The primary requirements of the service are thus: . To filter users with respect to a document (and conversely for the retrieval service) by using the three key concepts (competence model, zone of proximal development, and semantic reasoning) described previously. . To send a message containing information related to the document to the selected users (active dissemination) or to visualize information about the retrieved documents (intelligent retrieval). In addition to these primary requirements, the service has to: . . . . . .

Check in order to give access only to the employees of the company. Ensure that the dissemination is carried out by the administrators. Ensure that filtering is accurate and safe. Be configurable by the parameters of a configuration ontology. Be achievable on different platforms. Allow to visualize and navigate in the graph of ontologies for administrative purposes.

For the dissemination, the service must memorize the state of the documents in order not to omit documents in the list of documents to disseminate and not to importune the administrator by suggesting documents already disseminated. Filtering Strategy As evoked previously, the engine filters the users with regard to a given document (or conversely) by taking into account the necessary expertises to understand the document and those the user will acquire thanks to the document. The strategy uses not only the direct expertises (those explicitly related to the employee and the documentary resource), but also the indirect expertises obtained by semantically reasoning around the four concepts: role, level, competence, and skill:

Ontology for Knowledge Retrieval

857

. Role: When a person is assigned a role and if this role requires some expertises, then it is inferred that this person possesses the corresponding expertises. . Level: If a person has an expertise characterized by a level and a qualified competence, then the inference engine assumes that this person also has all the expertises, which are defined with a lower level and the same qualified competence. . Competence: If the competence associated to an expertise level is decomposed in several subcompetences, it is inferred that a person having the corresponding expertise also has all the expertises associated to the subcompetences with the same level and skill qualification. . Skill: If the skill qualifying a competence associated to an expertise level is more general than another skill, then it is inferred that each person having the corresponding expertise also has the expertise with the competence qualified by the less general skill at the same level. Considering all these criteria, the filtering system operates a closure on the expertises for a person and compares this closure to the expertise specifications for a document. The document is semantically in the proximal zone of development if the required expertises for the document are included in this closure and if at least one of the acquired expertises is not in this closure. Intelligent Querying Service The Intelligent Querying Service (IQS) is an advanced Semantic Webbased information retrieval service. It searches for documents for a user not only on the basis of concepts in the domain knowledge, but it takes also into account the working context and the user model. For example, a user can ask for documents that can help him to achieve a task. The service will provide only the documents that are related to this task and are relevant according to the competences required. A filtering mechanism analogous to the one used for the dissemination is used. In the case of this mechanism, only the required competences for a document must belong to the competences closure for a person.

CONCLUSION The MDKT project is aimed to provide enterprises with the means to allow them to address this knowledge acquisition and transmission problem related to technological innovation.

858

B. Lefebvre et al.

The challenges this project raises are numerous, including: . Defining a sufficiently explicit representation of the semantic content of the documents of the enterprise. From this representation, we should be able to characterize:
The degree of relevance for a human resource of elements of information appearing in the document.
The adequacy of the formulation of the content regarding the expertise of this resource. . Designing the most suitable set of tools to generate this representation in a partially automated way. . Setting up a user model and the associated competence model in order for the dissemination and retrieval tools to adequately permit:
The selection of people.
The selection of the suitable information.
A presentation of this information adapted to every category of resources. In order for the dissemination and retrieval services to be published on the Grid, an agreement must be found on the way knowledge about enterprises, employees, and competences, which are all needed by these services, must be represented. Shared and common ontologies implemented with the standard language proposed for the Semantic Web could adequately technically achieve this goal as shown by this project. The system, which has been built, has been instantiated in order to adequately test and validate the tools. However, in order to have an operational system in the context of a large enterprise and to help in the instantiation and maintenance task of the system, other tools have to be realized, which are crucial for the long-term life of the system. Among the remaining problems is the assistance to the domain ontology maintenance. This can be achieved by a statistical analysis of new documents. Such a solution has been considered and is under construction (Gargouri et al. 2003). REFERENCES Abecker, A., A. Bernardi, K. Hinkelmann, O. Kuhn, and M. Sintek. 1998. Toward a technology for organizational memories. IEEE Intelligent Systems 1340–1348. Abecker, A., A. Bernardi, K. Hinkelmann, O. Ku¨hn, and M. Sintek. 2000. Context-aware, proactive delivery of task-specific knowledge: The knowmore project. Int. Journal on Information Systems Frontiers (ISF) 2(3=4):139–162. Barry, R. E. 1998. Document management for the enterprise: Principles, techniques and applications. Journal of the American Society for Information Science 49(1):94. Berners-Lee, T., J. Hendler, and O. Lassila. 2001. The semantic web. Scientific American 284(5):35.

Ontology for Knowledge Retrieval

859

Fensel, D., F. van Harmelen, I. Horrocks, D. L. McGuinness, and P. F. Patel-Schneider. 2001. OIL: An ontology infrastructure for the semantic Web. IEEE Intelligent Systems 16(2):38. Fox, M. S., M. Barbuceanu, and M. Gruninger. 1996. An organisation ontology for enterprise modeling: Preliminary concepts for linking structure and behaviour. Computers in Industry 29(1=2):123. Gargouri, Y., B. Lefebvre, and J.-G. Meunier. 2003. Ontology maintenance using textual analysis. In Proceedings of the 7th World Multiconference on Systemics, Cybernetics and Informatics (SCI 2003), Orlando, Florida, USA. Horrocks, I. 2002. DAML þ OIL: A reasonable web ontology language. Lecture Notes in Computer Science 22872–22912. Nkambou, R. and B. Lefebure. 1996. A curriculum-based student model for intelligent tutoring systems. User Modeling 91–98. Paquette, G. 2002. Mode´lisation des connaissances et des compe´tences: Un langage graphique pour concevoir et apprendre. Sainte-Foy: Presses de l’Universite´ du Que´bec. Razmerita, L., A. Angehrn, and A. Maedche. 2003. Ontology-based user modeling for knowledge management systems. In Proceedings of the User Modeling Conference, pages 213–217, Pittsburg, Pennsylvania, USA, Springer-Verlag. Uschold, M., M. King, S. Moralee, and Y. Zorgios. 1998. The enterprise ontology. The Knowledge Engineering Review, Special Issue on Putting Ontologies to Use. 13. Vygotskii, L. S. and M. Cole. 1978. Mind in Society: The Development of Higher Psychological Processes. Cambridge, Mass.: Harvard University Press. Vygotskii, L. S. and A. Kozulin. 1986. Thought and Language. Translation newly rev. and edited by Alex Kozuli, 287. Cambridge, MA: The MIT Press.

NOTE 1. The concepts of the knowledge domain are also considered as instances in the data layer.