Visualizing a dynamic knowledge map using
Semantic Web technology
Hong-Gee Kim1,
Christian Fillies2, Bob Smith3 and Dietmar Wikarski4
1 Dankook Univ., Korea,
hgkim@dku.edu
2
Semtation GmbH, Schwabach, Germany
cfillies@semtalk.com
3 Cal. State University, USA
robsmith5@1talltrees.com
4 University of Applied Sciences Brandenburg, Germany
wikarski@fh-brandenburg.de
Abstract. Visual knowledge maps are being used to
improve the communication processes within global organizations. Knowledge maps are graphical presentations of
ontological knowledge as well as of business processes. Especially for enterprises working in a multi-cultural
space the explicit formalization of knowledge and business rules using graphical models seems to be a very promising
approach in order to improve discussion and learning processes. Publishing and automatic inference or search techniques are becoming available due to the latest standards for semantic web worked out by W3C. This article
gives an impression how to create end user interfaces for the ”Corporate Knowledge Base” using MS
Office and Visio with the modeling tool SemTalk. We are discussing several problems on capturing and maintaining large scale
knowledge bases. Specific attention is given to the problem of weighting and association of information from orthogonal ontologies, which
arises while using the same concepts in different graphical scenarios.
1 Graphical representation of a knowledge map
Knowledge management in an organization is the ability to
create business values and generate a competitive advantage. However, knowledge is not visible
in its nature so that managing it is very difficult. Tacit knowledge embodied
in the experiences of organizational members is easily lost unless it is transformed into a usable form. Knowledge mapping provides
a framework for visualizing knowledge that can easily be examined, refined, and shared by non-expert
knowledge users. A knowledge map can also be used as an interactive tool that links different
conceptualizations of the world.
Since people have their own concept of knowledge
in terms of its form and content, meaningful communication is difficult especially when the
number of communicative actors is large. There is a need to develop a common way of constructing and maintaining knowledge in a
visual form [8]. Many methods of knowledge representation have already been developed and devised for AI applications. Most of these techniques
are for
machine-processed and target specific systems. In contrast, the application discussed in this
paper provides a user-friendly method for developing a knowledge map that helps knowledge users visualize their
implicit ontologies and workflows. Knowledge about the
same object is represented differently depending on contexts. Since the
visualizing tool proposed in this paper suits with the dynamic feature of a
knowledge map, it helps people to modify and to combine ontologies across domains. SemTalk using semantic web
technology equips the user with a method for knowledge representation that is not only machine understandable but human readable
because it includes both graphical and textual forms of information.
Semantic nets are a powerful diagrammatic
knowledge representation technique. Figure 1 is an example of a knowledge map
that is represented in a semantic net. The knowledge mapping tool of SemTalk is more flexible and less constrained than semantic
network
systems in the sense that any graphical form of knowledge representation is possibly modeled including UML and Conceptual Graph. A
knowledge map represents meaningful relationships between concepts in the form of propositions. A proposition represented in a
knowledge map consists of two or more concepts linked by relational labels to form a semantic unit.
Fig. 1. A knowledge map of plants
2
Context dependency of a knowledge map
People conceptualize their world differently. Accordingly, a knowledge map about
the same object may contain different contents and structures depending on the
contexts for which they are generated. For example, a scientist usually has a view that electrical ‘current’
is a kind of “constraint-based events”, but in some contexts can share with
others the naïve view that it is a material substance. We can have multiple
views for a single concept depending on context [10].
As an object has a huge number of properties,
there are many ways of conceptualizing a given object, each serving a
particular goal. The concept ‘car’ may contain different information for a car dealer, a
manufacturer, a driver, and a cartoonist. We tend to conceptualize an object as having a
certain set of properties in the context of the kind of things involved. For example, there are explanatory networks for a car’s fuel systems, known
only to engineers, that consist of many mechanically defined terms unique to
engineers. A cartoonist could also have similar clusters of terms for the shapes and motions of cars. An object
is conceptualized and organized into ontology differently, yet some information can be shared across ontologies
when it is needed [11]. For example, although the cartoonist’s ontology of ‘car’ does not have
any mechanical information of a car, such information sometimes needs to be accessible to the cartoonist in a certain
situation. Figure 2 depicts how the two knowledge maps are merged in terms of
cross-ontological relationships.
Fig. 2. Ontology merge
Further information about merging of ontologies can found at [4].
3
Knowledge structure represented in a knowledge map
Knowledge maps are defined as “a graphical
representation of the connections by the brain in the process of understanding facts about something”
[7]. While the static side of knowledge mapping is to represent the connections
between properties identified during conceptualization, the dynamic side stands for the process of inferring
values on those properties in the problem solving or decision making contexts.
Concepts can be either directly or by analogy transferred from one domain to another. For example, the use of the
physical notion of transitory state can be transferred into the domain of certain business
management problems. The dynamic aspect of knowledge mapping is used to improve
the communication processes within global organizations. Especially for enterprises working in a multi-cultural space, the explicit formalization of knowledge and business rules using
knowledge maps seems to be a very promising approach in order to improve communication and learning processes.
The actual knowledge does not have a static
structure but is dynamically constructed by identifying and indexing pieces of information or knowledge components depending
on contexts. Figure 3 describes how knowledge is represented in a knowledge map
that shows the hierarchical structure of knowledge. Understanding is not just knowing an
item of knowledge, but knowing how the supporting knowledge relates each higher knowledge item [8]. A measure of
importance
represents how important each supportive piece of knowledge item is to the higher one. The weighting in
the association between knowledge components can vary depending on contexts. In
the actual use of a knowledge map granularity is flexibly applied in the sense
that a certain knowledge component, in this example ‘cooking chicken’, may
consist of a deeper level knowledge map with a greater granularity.
Fig. 3.
Knowledge structure about cooking chicken
4 A
real world
use case
Large and successful organizations today can afford to invest resources into formalizing corporate ontologies, but medium sized organizations can seldom afford the time and the resources to effectively execute these important projects.
Our use case is a 105 year old engineering and chemical testing lab
employing over 100 people. The engineering and construction professionals have conflicts with the chemistry lab
professionals. The IT department is very small and under-funded. They use WinWord and Excel on a network, and have used Visio for planning and training purposes in the
past. SemTalk now offers a low cost approach to visualizing workflows and their implicit ontologies
based on the W3C notation RDFS using Visio shapes which are relevant to their
business problems.
Fig. 4.
Knowledge structure for the chemistry testing lab
An explicit examination of each group's ontology
(and
specialized jargon) significantly enhances the CEO/Owners' ability to more effectively balance
resources within the organization. Applying an importance measure on the specific issues found in both organizational sub-structures
helps to make communication deficits explicit. As a direct result a 20%
increase in profit 6 months after project completion is expected.
5
SemTalk
SemTalk is being used in ontology projects
helping people to agree on a common language. As described in Fillies et al.
[5] those graphical ontologies may be used in several ways such as terminology
control for technical writers. Ontologies represented in an application independent XML
based format are an important building block for any knowledge management system, for business process modeling and for the consistent definition of
large projects e.g. using MS Project. Ontology based business process models can be maintained,
translated and reused with significantly less effort than conventional process
models. This especially applies for process models describing web services.
5.1 Architecture of SemTalk
SemTalk does work on an RDF(S)-like XML data structure. Diagramming information and object oriented features like methods and states have been added to
RDF(S). It also has an optimized structure for basic inferences as inheritance and graph traversals. There is an object engine providing a COM API in order to be able to use the engine
within MS Office products. For the graphical presentation of models we have used MS Visio for two reasons: (i) the tool is widely used in
industry, therefore people are used to it and (ii) it is easily extensible through an API.
Fig. 5.
Architecture of SemTalk
The SemTalk object engine is used to define
semantics - in other words a Meta Model - for existing Visio shapes. You can graphically define which shapes are
allowed to be connected with each other. SemTalk supplies the infrastructure to
define complete modeling methods inside Visio. Those methods are e.g. for DAML [3], for Enterprise Resource Planning (ERP) product modeling and for Business Process Modelling
(BPM) methods. SemTalk has a couple of interfaces to CASE tools like Rational Rose and to BPM tools. There is a simple report generator that creates HTML tables by
using XSL for formatting.
5.2 Notation for Semantic Nets
In respect to the very broad audience we want
people to be able to read our models without learning a notation. We have best
experiences using the very simple bubble notation, shown in some of the
pictures below. It is important to label most of the links and not to use graphical encodings which are known from graphical
languages as Entity Relationship diagrams.
For readers with a technical background more complex notation with
various shape types can be used. Examples are the DAML Notation and e.g. a user interface for a product configuration
engine.
One of the great advantages of using Visio is
that is contains a large collection of predefined and extendable shapes. The
shapes correspond quite natural to classes. Using pictures improves the
acceptance of the models which is an important success factor in Knowledge Management.
5.3 Referencing external Knowledge Bases
WordNet®, which was developed by the Cognitive Science Laboratory at Princeton University
under the direction of Professor George A. Miller, is a huge online lexical reference system whose design
is inspired by current psycholinguistic theories of human lexical memory. English nouns, verbs, adjectives and adverbs are organized into synonym sets,
each representing one underlying lexical concept. Different relations link the
synonym sets. SemTalk uses WordNet via Dan Brickley’s RDF(S) web service for WordNet 1.6 [2] [13].
Fig. 6. A
small vehicle model built from WordNet
The models are being built with external model
repositories incrementally. Once you have used a class name in a model you
can look for related objects in external repositories and integrate them into your
model (Figure 6). The idea of using an external glossary basically ensures that
people are talking about the same thing with a well defined Uniform Resource Name (URN) to
identify objects and related hyperlinks to access their definitions. The other benefit
users have from such ontologies is that they are getting hints for related objects or subclasses to use in the
model.
The objects remember their origin and can be refreshed (or replicated) from their
external data source once the source has changed. In a very similar way you can
link one class to another class living in an external model which was created
using SemTalk and which is published on a web server. This technology results in a
web of hyperlinked models based on RDF(S) as a common standard.
6
Weighted Knowledge Maps in SemTalk
Cross ontology integration is a very common
problem which arises as soon as multiple organizational units such as different departments within one company
are involved. It becomes very important once business partners with a very inhomogeneous cultural
background and communication strategies are being forced to solve real world problems together. This is
especially the case if corporations from Asia move to western markets or vice versa.
Abstract and graphical models for knowledge and business processes have been used from the very first days of
mankind to ease communication. SemTalk is a modeling tool designed to create
knowledge structures in the semantic web format RDFS (Resource Description Framework, [1]. The semantic web is a
kind of a distributed world wide knowledge model. One of the basic ideas of the semantic web
is to denote concepts of discourse by URN. Once a group of users have has
agreed that they are talking about the same topic, they can refer to it from
their specific application models by a public URN. This technique is used to disambiguate words by explicitly mentioning
homonyms and assigning synonyms to concepts. Beyond using URNs and synonyms, SemTalk relies on
the use of manually clustering of information on diagrams, contexts or scenarios. Each object and a subset of its associations can participate in multiple scenarios.
The technique of weighting the importance of an association in a specific context offers first of all
additional information. For larger projects the importance factor of associations helps to reuse the object in contexts build by
different people, because a statement made in one context may be less important in other contexts.
Fig. 7.
Using line width to visualize importance
A very simple but effective way to visualize importance is to use graphical
properties such as line width or no node size in order to emphasize specific aspects of the scenario.
Adding weighting and importance factors to the RDFS class model was
possible because of two reasons: 1. RDFS is based on XML and 2. from the tool builders point of view
because SemTalk has an open meta model, which allows the extension of the association (RDF speech:
“Property”) and regards them as first class objects.
SemTalk is integrated in MS-Office. It has a
Visio based graphical user interface which makes it easy to use for a broad range of users.
Using Office XP SmartTag technology, semantic web glossaries can be used from
all MS Office applications to lookup words in an ontology or process model.
7
Future Research
In this paper we have shown how to apply dynamic
knowledge maps to Semantic Webs mainly for the purpose of improving communication and understanding between human readers of
models. The semantic Web also has an important influence for the communication of programs or machines. Interpreting process descriptions by workflow engines or executing processes with MS Project having a (fuzzy)
measure of “importance” has to be investigated.
References
1. Berners-Lee,
T. Hendler, J., and Lassila, O.: published an article about the Semantic Web in Scientific
American. "A new form of Web content that is meaningful to computers will unleash a
revolution of new possibilities", See http://www.scientificamerican.com/2001/0501issue/0501berners-lee.html
2. Brickley,
D.: RDF(S) web service for WordNet 1.6, cf. http://xmlns.com/2001/08/wordnet/
3. Darpa
Agent Markup Language (DAML): cf. http://www.daml.org
4. Doan,
A., Madhavan, J., Domingos, P., Halevy, A.: “Learning to Map between Ontologies
on the Semantic Web”, WWW2002, May 7-11, 2002, Honolulu, Hawaii, USA ACM
1-58113-449-5/02/0005
5. Fillies,
C., Wood-Albrecht, G., Weichhardt, F.: A Pragmatic Application of the Semantic Web
Using SemTalk, WWW2002, May 7-11, 2002, Honolulu, Hawaii, USA ACM
1-5811-449-5/02/0005
6. Flynn, J.: DAML Visio Shapes, cf. http://www.daml.org/visiodaml/
7. Gomez,
A., Moreno, A., Pazos, J., Sierra-Alonso, A.: Knowledge maps: An essential technique
for
conceptualization, Data & Knowledge Engineering 33. (2000) 169-190
8. Gordon, J.L.: Creating knowledge
maps by exploiting dependent relationships, Knowledge-Based Systems 13. (2000) 71-79
9. Gruber: http://www-ksl.stanford.edu/kst/what-is-an-ontology.html
(2001)
10. Kim, H.-G.: “A Psychologically
Plausible Logical Model of Conceptualization.” Minds and Machines, 7. (1997) 249-267
11. Kim, H.-G.: “Formalizing Perspectival
Defeasible Reasoning.” Proceedings of the 30th Hawaii
International Conference on System Science, Vol. V. (1997) 347-353
12. W3C: RDF Schema Specification. http://www.w3.org/TR/PR-rdf-schema/,
1999, O. Lassila and R. Swick: Resource description framework (RDF), model and syntax specification
Technical report, W3C, 1999, W3C Recommendation http://www.w3.org/TR/REC-rdf-syntax
13.WordNet: cf. http://www.cogsci.princeton.edu/~wn/