Research and Advances
Architecture and Hardware Ontology applications and design

A Collaborative Approach to Ontology Design

Creating a general ontology characterizing the conduct of knowledge management.
Posted
  1. Introduction
  2. Approaches to Ontology Design
  3. Ontology Design via Collaboration—A Case Study of a Delphi Approach
  4. Conclusion
  5. References
  6. Authors
  7. Figures
  8. Tables

Because of the importance of ontologies in the emergent era of knowledge-based organizations, better support for their creation is needed. Designing a substantial ontology is not trivial, and designing it so it has relevance and value to a broad audience is even more challenging. Generally, ontology is a branch of philosophy dealing with the order and structure of reality. Here, we adopt Gruber’s view that an ontology is an explicit specification of an abstract, simplified view of a world we desire to represent [2]. It specifies both the concepts inherent in this view and their interrelationships. A typical reason for constructing an ontology is to give a common language for sharing and reusing knowledge about phenomena in the world of interest.

For instance, an ontology for knowledge-based organizations would furnish terms and definitions for concepts (entities, objects, events, processes, goals, and results) deemed to be important in characterizing these organizations at a desired level of detail; concepts and their relationships would be further characterized in terms of axioms and constraints that may be expressed more or less formally. Among those who adopt the ontology, its terms are used in asking and answering questions, making assertions, offering insights, describing practices, and discussing investigations pertaining to the conduct of knowledge management (KM).

In building and applying an ontology, it is important to clearly make the following distinction: on one hand, there is the ontology itself, which specifies concepts used in a domain of endeavor, concepts whose existence and relationships are true by definition or convention. On the other hand, there are empirical facts about these concepts and relationships. They are not part of the ontology, although they are structured by it. They are subject to context, observation, testing, evaluation, or modification. In the domain of financial services, for example, concepts such as currency, equities, trade execution, and trade settlement are parts of an ontology. The fact that three days elapse between trade execution and trade settlement in the U.S. context is an example of knowledge that is not part of the ontology, whose effectiveness can be assessed, and which can be modified.

Ontological commitment is important. It is the agreement by multiple parties (persons and software systems) to adopt a particular ontology when communicating about the domain of interest, even though they do not necessarily have the same experiences, theories, or prescriptions about that domain. For instance, all financial services practitioners agree that trade execution and trade settlement exist and that execution precedes settlement. However, there may be disagreement about whether the elapsed time should be three days or five days. Where ontological commitment is lacking, it is difficult to converse clearly about the domain and benefit from knowledge of others. It follows that development of an ontology should proceed with an eye toward ensuring that its potential users will find its characterizations to be sufficiently complete, correct, clear, and concise. Working toward ontological commitment should not be an afterthought, but rather an integral aspect of ontological engineering. This contention underlies the collaborative approach to ontology design we advocate.

Back to Top

Approaches to Ontology Design

Ontological engineering is concerned with the principled design, modification, application, and evaluation of ontologies. The accompanying table outlines five approaches to ontological design: inspiration, induction, deduction, synthesis, and collaboration. These may be used in the initial design of an ontology or the modification of a design (for example, in reaction to feedback on its applications, evaluations of its features, or domain changes). Hybrids of the approaches are possible.

Inspirational approach. With an inspirational approach to ontology design, a developer starts from a premise about why an ontology is needed. Using individual imagination, creativity, and personal views on the domain of interest, he or she proceeds to design an ontology that aims to meet the recognized need. For instance, the notion of a decision support system (DSS), as distinct from a management information system, began to emerge in the 1970s. Recognition of the need (and opportunity) for such a system to embody database management, operations research, and artificial intelligence methods in an integrated fashion inspired creation of the generic DSS framework. This is an ontology whose concepts were created to accommodate the representation and processing of descriptive, procedural, and/or reasoning knowledge within any system devised for supporting decision-making [10].

Unless a developer has the ability to dictate adoption of the ontology (when building a community of intelligent agents), ontological commitment for the final product of inspirational ontology design may be narrow. Exceptions occur when the developer’s personal views are aligned with those of many potential adopters or the created specification strikes a responsive chord with a large audience. Evidence of this will be seen in the extent to which the ontology is used, its longevity, and the degree to which it leads to more extensive ontologies that adopt/adapt its basic features.

Inductive approach. With an inductive approach, an ontology is developed by observing, examining, and analyzing a specific case(s) in the domain of interest. The resulting ontological characterization for a specific case is applied to other cases in the same domain. For example, based on an examination of KM-related behaviors and processes in a specific organization, the developer advances an ontology intended to account for the observed phenomena. The degree of commitment to this ontology is seen in how widely it is applied to the study or administration of other organizations. Can an ontology based on the KM activity in one consulting firm be usefully applied in understanding KM behaviors and processes in other consulting firms, or more broadly, in financial services or manufacturing firms?

Deductive approach. Conversely, a deductive approach to ontology design is concerned with adopting some general principles and adaptively applying them to construct an ontology geared toward a specific case. This involves filtering and distilling the general notions so they are customized to a particular domain subset. It can also involve filling in details, effectively yielding an ontology that is an instantiation of the general notions. For example, one could take the general DSS framework as a starting point in deducing an ontology customized for characterizing text-based, solver-based, or rule-based decision support.

Synthetic approach. With the synthetic approach, a developer identifies a base set of ontologies, no one of which subsumes any other. The traits of these base ontologies, perhaps along with other selected concepts pertaining to the phenomenon being described, are synthesized to develop a unified ontology. Because it embraces multiple ontologies, the result may be prone to adoption by its adherents and an opportunity presented for them to interact in a coherent fashion.

For instance, various KM ontologies have been advanced; they have some overlaps, but no one of them includes all the concepts of any other [4]. Some are fairly narrow, describing KM phenomena for a specific organization (such as a consulting firm) or focusing on one aspect of KM (such as knowledge transfer); others aim to be more general. Synthesizing a unified ontology from all these involves systematic integration of their concepts, elimination of sketchier characterizations in favor of more fully developed ones, and reconciliation of different terminologies.

Collaborative approach. With a collaborative approach to ontology design, development is a joint effort reflecting experiences and viewpoints of persons who intentionally cooperate to produce it. Chances for relatively wide acceptance are enhanced if these persons are diverse in the contributions they make. This helps reduce blind spots in the ontology and enrich its content. On the other hand, coordination of the design process may suffer if too many persons are directly involved. The process itself could range from being strongly anchored, with a proposed ontology as a starting point for iterative improvements, to comparatively unstructured serendipitous discussion. In order to execute a collaborative approach, a consensus-building mechanism needs to be employed.

In the following example of a collaborative approach, participants are a sizable, balanced mix of scholars and practitioners, all of whom are experienced in the domain and who represent diverse viewpoints, experiences, and backgrounds. This approach uses a Delphi-like method to structure collaboration in the direction of consensus: an initial ontology (produced via synthesis) is critiqued by participants; the developer revises the ontology, attempting to address participants’ critiques; participants critique the revised ontology, which forms a basis for another revision, and so forth, until participants collectively agree on an ontology.

There are pros and cons to the five approaches. The inspirational approach can be questioned as lacking theoretical underpinning and may be impractical. However, this approach may also yield unique, innovative ontologies. An ontology designed via the induction approach may fit a specific case, but may not be generalizable. The deductive approach presupposes existence and selection of an appropriate scheme of general characterizations from which an ontology for a specific case can be devised. The synthetic approach implicitly covers the first three approaches in that its base set could involve any or all of them, and the base set can be supplemented by constructs currently existing in the literature. However, this approach is interpretative in nature, relying heavily on developer synthesizing skills. None of the first four approaches has a built-in evaluation facility to assess quality/acceptability of the resultant ontology. In contrast, the collaborative approach relies on assessments from diverse vantage points and tends to build commitment by iteratively reducing participants’ objections. This approach relies heavily on the nature of participants, the degree of their involvement/diligence, and developer skills in overseeing the collaborative process.

Back to Top

Ontology Design via Collaboration—A Case Study of a Delphi Approach

The figure demonstrates a Delphi-oriented structuring of collaboration in the context of an actual case where we used it to develop an ontology for the conduct of KM in organizations. Descriptions of the resultant ontology appear in [5–7]. The basic phases in this case can serve as a guide for other situations where a collaborative approach to ontology design is desired. As the figure shows, these phases are preparation, anchoring, iterative improvement, and application.

Phase 1. In the preparatory phase, we defined design criteria for the ontology, specified boundary conditions for the ontology, and determined a set of standards for evaluating the ontology. The design criteria we selected for the KM ontology were comprehensives, correctness, conciseness, clarity, and utility. These criteria were important in both guiding development of the ontology and assessing the degree of its success.

We applied the design criteria within three boundary conditions: business boundary, descriptive boundary, and detail boundary. That is, our focus was on KM in business organizations, on describing KM phenomena rather than prescribing KM methodologies, and proceeded in a top-down fashion to a depth of at least two levels.

To help keep design activity on course, we predetermined standards against which the ontology was evaluated as it developed. These included a set of existing KM ontologies, concepts evident in KM “best practices,” notions inherent in issues raised by KM researchers, and traits identified via cases and surveys focusing on KM in organizations. The effort to achieve consistency with these standards shaped the ontology as it evolved in the next two phases. It was also shaped by critiques and suggestions during the collaborative third phase.

Phase 2. Any of the approaches in the table could be used to produce an ontology for “seeding” the collaborative design activity, serving as an anchor to help focus the attention of collaborators. We chose to synthesize an ontology by consolidating, organizing, and integrating languages, concepts, and their interrelationships found in the evaluation standards. This anchor ontology evolved through several iterations. At the completion of each iteration, we evaluated the ontology in terms of design criteria and relatively to standards. Modifications were made until we were satisfied that design criteria were met with respect to the evaluation standards.

The synthetic approach is interpretive. Even though our aim was to meet design criteria, the anchor ontology may not fully satisfy them. Moreover, it is possible the evaluation standards may not collectively cover all key concepts in this domain. The point of the next phase is to ascertain where the ontology falls short, revise it to overcome those shortcomings, and obtain independent evaluations of the revised ontology.

Phase 3. The collaborative approach involves an adaptation of the Delphi method [11]. This is a formal technique for collecting and integrating the views of multiple persons about some topic. Each participant independently provides views in writing to a leader, who prepares a document reflecting the combined views as feedback for the next round. In the second round, participants furnish their independent written views in light of the feedback. In this way, the leader attempts to foster a convergence of views across successive rounds.

The Delphi technique gives a systematic way for gathering perspectives and critiques on an ontology as a basis for iterative improvement. Secondarily, it gives a convenient way to gather independent assessments of the ontology with respect to the design criteria. We sent the anchor ontology and a questionnaire to members of a panel for their critique and comments. As the developer, we organized and analyzed their responses as a basis for ontology revision. A second round commenced with our production of a revised ontology which, along with a summary of first-round results and a questionnaire, was sent to panelists for critique and comment. This process iterated until panelists regarded the ontology as successful with respect to design criteria.

The panel was comprised of over 30 persons from four continents, with varied backgrounds, and evenly divided between those with KM track records in business and academia. We devised a questionnaire for structured elicitation of critiques in terms of the design criteria. Panelists’ perceptions were captured using Likert-scale items, as well as written responses. When dissatisfied with an aspect of the ontology, a panelist was asked to elaborate on why and suggest improvements.

Participants’ responses were recorded in a database from which a document of their critiques and comments was generated. Responses were grouped by questionnaire item. For each item, we reviewed and evaluated all comments and critiques, separating them into two groups: those for further consideration and those beyond boundary constraints. We classified responses in the first group into three categories: concerns that were repeated and/or seemed of major importance; concerns that were not so frequent and/or major; and concerns that were infrequent and incidental. Next, we prepared a summary document, organizing responses into these categories for each questionnaire item and providing relative frequency distributions for Likert-scale responses. This summary document guided and stimulated our modifications of the ontology in the next round.

Revisions were classified as fundamental, new additions, and clarifications. Fundamental revisions involved extensive effort, developing specifications of new concepts that emerged from discussion stimulated by participants’ comments, detailing and further characterizing concepts existing in the critiqued ontology, and further justifying inclusion of ontology elements. New additions involved adding suggested elements and describing their natures and relationships with other ontology elements. Clarification occurred when an element was already in the ontology, but needed to be emphasized or explained more clearly. In the second round, each panelist received a paper describing the revised ontology, a questionnaire, and response summary document. Second-round responses were analyzed in the same manner as first-round responses. The iterations terminated when the panelists agreed the ontology was successful with respect to design criteria and no major concerns remained.

Phase 4. In this phase, we explored ontology utility by applying it in various ways: demonstrating it provides a unifying view of KM phenomena [8]; using it as a framework to explore knowledge selection processes and technologies [3]; and generating the knowledge chain model of competitive analysis [9, 12].

Back to Top

Conclusion

Collaborative ontology design can be applied to produce ontologies for e-commerce, agent coordination, distance learning, virtual enterprises, and other phenomena. Here, we have provided guidelines and a sample for this kind of ontology development. Beyond this, experience gained from the KM-ontology design case offers lessons that should be considered when using the collaborative design approach.

There can be panelist attrition across rounds. Even though every panelist from round one indicated willingness to participate in round two, some did not. The task of reviewing a substantial ontology and carefully answering questions is time-consuming. Longer, more flexible response turnaround may help minimize attrition.

In some cases, a respondent commented that improvements were made across rounds, but comparison of Likert-scale items across rounds did not reflect this. One possibility for improving response consistency is to provide each panelist with a questionnaire that shows his or her responses to the same items in the prior round.

There was a tendency for the ontology description to grow across rounds. More detail and elaboration were requested by first-round panelists, and handling new ontology elements that emerged from critiques required additional specification. If ontology specification becomes very lengthy, it can hike the attrition rate among volunteer panelists. Our objective was to keep the specification under 20 pages. However, it grew well beyond that. One way to manage the length is to divide the ontology into less formidable parts, using a separate Delphi process (perhaps with a separate panel) for each.

KM is a dynamic and multidisciplinary subject. As such, it presents something of a moving target for panelists and developer alike. Practical research boundaries, sound design criteria, and diverse evaluation standards are important for coping with ontology design for dynamic subject areas. Dynamic ontology development envisioned in work on the semantic Web may well develop into a useful approach for ontological engineering of dynamic domains of interest [1].

One panelist observed that the Delphi process can involve an anchoring effect. Indeed, the ontology devised in the second phase serves as a focal point anchoring the whole process. While this helps structure the collaborative effort, it may also restrict panelists’ expression and imagination. Thus, we advocate the use of synthesis with a broad, inclusive base set when preparing the anchoring ontology.

Aside from the final ontology, some panelists volunteered that participation in the collaborative design process was itself very useful. The summaries of panelists’ responses enabled them to gain new insights and exposed them to diverse perspectives. As one (a practitioner) commented: “The Delphi process—in particular reading the summary of participants’ comments—has been very rewarding and I thank you for it and wish to continue.” Another wrote, “It is thought-provoking work and I have enjoyed pondering it.”

Collaborative ontology design promotes ontological commitment and provides a mechanism whereby panelists share and exchange their perspectives and expertise. Panelists are active participants in developing the product (the ontology) rather than passive recipients. Their involvement in offering critiques and suggestions helps in developing, refining, and modifying their own work related to a domain of interest.

Back to Top

Back to Top

Back to Top

Figures

UF1 Figure. A collaborative approach to ontology design.

Back to Top

Tables

UT1 Table. Approaches to ontology design.

Back to top

    1. Berners-Lee, T., Hendler, J. and Lassila, O. The semantic Web. Scientific American (May 2001), 28–37.

    2. Gruber, T.R. Toward principles for the design of ontologies used for knowledge sharing. Int. J. Hum. Comput. Stud. 43, 5/6 (1995), 907–928.

    3. Holsapple, C.W. and Joshi, K.D. Knowledge selection: Concepts, issues, and technology, in J. Liebowitz, Ed., Handbook of Knowledge Management. CRC Press, Boca Raton, 1999, pp. 7–22.

    4. Holsapple, C.W. and Joshi, K.D. Description and analysis of existing knowledge management frameworks. In Proceedings of the Hawaiian International Conference on System Sciences. Maui, January, 1999.

    5. Holsapple, C.W. and Joshi, K.D. An investigation of factors that influence the management of knowledge in organizations. J. Strat. Info. Syst. 9, 2-3 (2000), 235–261.

    6. Holsapple, C.W. and Joshi, K.D. 2001. Organizational knowledge resources. Decision Support Systems 31, 4, 39–54.

    7. Holsapple, C.W. and Joshi, K.D. Knowledge manipulation activities: Results of a Delphi study. Information and Management. To be published.

    8. Holsapple, C.W. and Joshi, K.D. The evolution of knowledge management frameworks, in S. Barnes, ed. Knowledge Management Systems: Theory and Practice. International Thomson Business Press, London, 2002.

    9. Holsapple, C.W. and Singh, M. The knowledge chain model: Activities for competitiveness. Expert Syst. Apps. 20, 1 (2001), 77–98.

    10. Holsapple, C. and Whinston, A. Decision Support Systems: A Knowledge-Based Approach. West, St. Paul, MN, 1996.

    11. Lindstone, H. and Turoff M. The Delphi Method: Technology and Applications. Addison-Wesley, Reading, MA, 1975.

    12. Singh, M. Toward a knowledge management view of electronic business: Introduction and investigation of the knowledge chain model for competitive advantage. Ph.D. dissertation, University of Kentucky, Lexington KY, 2000.

Join the Discussion (0)

Become a Member or Sign In to Post a Comment

The Latest from CACM

Shape the Future of Computing

ACM encourages its members to take a direct hand in shaping the future of the association. There are more ways than ever to get involved.

Get Involved

Communications of the ACM (CACM) is now a fully Open Access publication.

By opening CACM to the world, we hope to increase engagement among the broader computer science community and encourage non-members to discover the rich resources ACM has to offer.

Learn More