.
K N O W L E D G E
M A N A G E M E N T
Using AI in Knowledge
Management: Knowledge
Bases and Ontologies
Daniel E. O’Leary, University of Southern California
C
ONSULTING AND PROFESSIONAL
services firms are often among the first organizations to adopt a new technology. The
reason is clear: they’ve got to know the technology before their clients do.1 Consequently,
there appears to be a technology life cycle in
consulting firms, where the firms first try the
technology for internal use, before selling that
same technology to their clients in the form of
consulting engagements. We see this process
at work with knowledge-management systems and their two main components: knowledge bases and ontologies.
Knowledge-management systems employ
a wide range of knowledge bases, especially
including best-practices knowledge bases. To
use those knowledge bases effectively, the
consulting firms must be able to generate
ontologies that allow users to pinpoint what
resources they need and want. Ontologies and
knowledge bases are closely related in knowledge management. Ontologies define the
knowledge base’s characteristics and views,
while also employing models that are helpful
in knowledge-base definition and access.
This article looks at the way knowledge
bases interact to form effective knowledgemanagement systems, and in particular at the
way leading consulting firms such as Price
Waterhouse, Ernst & Young, and Arthur
Andersen apply those systems to their businesses (see the “Consulting firms” sidebar).
34
THIS ARTICLE LOOKS AT THE USE OF ARTIFICIAL
INTELLIGENCE IN KNOWLEDGE-MANAGEMENT SYSTEMS,
FOCUSING ON SUCH AI-RELATED TECHNOLOGIES AS
KNOWLEDGE BASES AND ONTOLOGIES. BECAUSE THESE
TECHNOLOGIES BOTH DEPEND ON PARTICULAR SETTINGS, THE
AUTHOR DISCUSSES KNOWLEDGE MANAGEMENT AS PRACTICED
AT THREE MAJOR PROFESSIONAL SERVICES FIRMS.
What is knowledge
management?
Knowledge management is the formal
management of knowledge for facilitating
creation, access, and reuse of knowledge,
typically using advanced technology.
Knowledge-management systems contain
numerous knowledge bases, made up of
numeric and qualitative data (searchable Web
pages, for example). In addition, knowledgemanagement systems often allow discussion
groups that focus on a single set of issues or
a specific activity, such as particular software
or a single consulting engagement.
There are many similarities between AI
and knowledge management. For example,
knowledge-management systems employ
1094-7167/98/$10.00 © 1998 IEEE
knowledge bases, but for both human and
machine consumption. As elsewhere in AI, a
knowledge-management system also depends
on ontologies to facilitate communication
between its multiple users and links between
multiple knowledge bases. In turn, knowledge
bases rely on ontologies for unambiguous
specification of views and structure.
What do knowledge bases do?
Knowledge bases contain the content of
the knowledge-management system.
Knowledge bases in consulting firms.
Knowledge bases usually depend on the specific business and domain in which the orgaIEEE INTELLIGENT SYSTEMS
.
Consulting firms
nization is engaged. Consulting firms typically have knowledge bases that include proposals, engagements, best practices, and a
wide range of other topics as a part of their
knowledge-management systems. Engagement knowledge bases summarize information about different jobs that are captured in
working papers, either actual or virtual. Proposal knowledge bases capture information
about proposals that the particular firm has
made to generate engagements. News knowledge bases provide news releases of interest
to the consulting firms, while other knowledge bases allow access to recent journal and
magazine articles. Best-practices (or leading
practices) knowledge bases provide access
to enterprise processes that appear to define
the best ways of doing things. Expert knowledge bases identify who in the firm is expert
in a particular set of activities.
Knowledge-base development. Not only do
knowledge bases differ in their content, but
also in their development complexity—or the
difficulty of developing the knowledge base,
relative to other knowledge bases. A number
of factors influence this difficulty.
First, some knowledge bases capture information that was generated using a limiting
technology, such as paper. Consultants have
often kept paper-based knowledge bases,
including engagement or proposal databases,
where the paper format limited access. However, despite these limitations, those consultants did manage to capture the knowledge,
so many of the difficulties of capturing the
knowledge have already been addressed.
Accordingly, knowledge about previously
captured information is likely to be better
understood than knowledge that has never
been captured. Therefore, the fact that a
knowledge base has or has not been developed previously—independent of storage
format—helps knowledge managers gauge
the difficulty of developing that particular
knowledge base.
Also, many knowledge bases use a single
source of information—engagements or proposals, for example. Because these knowledge bases are limited to a single type of
information, each type of knowledge base
generally is easier to develop and maintain
than knowledge bases with new knowledge
and multiple sources of knowledge. That’s
because there is no need to integrate multiple
databases or search for information in multiple locations. Knowledge bases developed
from a single source were among the first
MAY/JUNE 1998
used. For example, at the consulting firm of
Arthur Anderson: “Since the early 1960s,
there was a resource known as the Subject
Files ... which allowed someone who was a
recognized expert in a subject...who wrote a
‘white paper,’ and this was filed and indexed
into ‘Subject Files.’”2
Furthermore, knowledge bases can use
information from both internal or external
sources, although the quality and quantity of
external information generally is less predictable and less controllable than internal
information. Also, multiple-source knowledge
bases derive from decisions based on both
acquisition and integration. Consequently,
those knowledge bases that require both types
of information are more difficult to generate
and maintain easy access to than those that
gather information from a single source.
Finally, some databases use virtually all the
information available on a source document,
such as accounting documents and, in particular, purchase orders. However, in such other
databases as best-practices databases, the
information must be abstracted, synthesized,
or integrated with other information. Development complexity of the knowledge base
generally increases as the data is manipulated.
Best-practices knowledge bases. These
knowledge bases are particularly complex
and difficult to develop. As noted by Arthur
Andersen:
We underestimated the sheer effort necessary to
translate ... knowledge about best practice into
useful explicit knowledge. The central team
could not, on its own, extract the ... knowledge
of the consultants and the professionals in the
field.... After a significant effort, the team had
produced a CD-ROM with the classification
scheme, but only 10 of the 170 processes populated, and with limited information. Further, the
information was not actionable—it added little
to those with deep knowledge of the area, and
Content
identification
Arthur Andersen is headquartered in
Chicago and Geneva, Switzerland, with 370
offices in 90 countries. They have approximately 45,000 employees worldwide. Revenues are more than $5 billion. They focus
on accounting, auditing, and tax services;
business consulting; and economic and
financial consulting.
Ernst & Young is headquartered in Cleveland and New York and has roughly 75,000
employees worldwide. They focus on assurance and advisory business services, management consulting, and tax services.
Price Waterhouse is headquatered in New
York and has over 100 offices in 118 countries. They offer accounting, auditing, tax,
and consulting services to companies and
government agencies. They have KnowledgeView Centers in Dallas and London.
was not enough to help those who had less experience....The initial offering almost died an early
death—it seemed much effort for little payoff.3
To understand the reasons for complexity
and difficulty of developing best-practices
knowledge bases, let’s look at Ernst & Young’s
model of a development environment for their
leading-practices knowledge base.4 Figure 1
shows that best-practices databases draw
information from multiple sources, both inside
and outside the firm. This content must be
ranked, abstracted, synthesized, and reviewed.
Best-practices knowledge bases fit each of
the criteria of difficulty I laid out earlier.
Best-practices knowledge bases are a recent
development. There is little evidence of summaries of best-practices knowledge bases
prior to the systems we’re discussing here.
Arthur Andersen claims that it first started its
Global Best Practices knowledge base in
1991.2 As Figure 1 shows, enterprises must
search out information from many sources to
Internal
information
Document repositories,
engagements
Abstract and
synthesize
Populate
External
information
Journals, articles, studies
Review
Figure 1. Ernst & Young’s leading-practices knowledge-base development environment.
35
.
generate a best-practices database and, because there are no generally available sources
of best-practices information, they must
search them out as well. Because best practices are constantly changing, enterprises
also must constantly change their knowledge
bases. They must therefore take a more active
approach to developing best-practices knowledge bases than many other types of knowledge bases. As a result, for consulting firms,
the existence of best-practices knowledge
bases signals the extent of development of a
knowledge-management system: less developed knowledge-management systems generally do not have best-practices databases;
more developed systems do.
What are best-practices
knowledge bases?
These databases capture information and
knowledge about the best way to do things.
They capture knowledge about processes,
rather than artifacts. A wide range of enterprises have used best-practices knowledge
bases. General Motors-Hughes Electronics
captured best-process reengineering practices in a database,5 and major consulting
firms such as the three we are discussing in
this article have also have developed bestpractices knowledge bases.
Arthur Andersen’s KnowledgeSpace. Figure 2 shows Arthur Andersen’s Global Best
Practices knowledge base. Within GBP,
Arthur Andersen defined the types of knowledge to be included in the knowledge base:
best practices by process, process definitions,
examples of companies applying the process,
relevant engagements by process, internal
experts, performance measures, presentations, and studies and articles.
Recently, KnowledgeSpace (http://www.
knowledgespace.com) became available as a
service over the Internet to subscribers. In
addition to having access to best-practices
information, subscribers now can access
news, discussion groups, and other resources.
Subscribers currently can access a small percentage of the best-practices knowledge base
and the basic framework. Materials include
a definition of the process, an executive summary, and a questionnaire that users can fill
out to provide information regarding the particular processes at their firms. Questions
take a yes-no format—for example, “Managers of purchasing departments accept
36
accountability for its performance?”—suggesting that a best practice relates to acceptance of accountability.
Ernst & Young’s Leading Practices KBase. For its own internal use, Ernst & Young
has developed a knowledge base of over
5,000 leading practices, which it has deployed in over 30 countries. The knowledge
base contains information from five sources:
benchmarks (quantitative values of performance measure by industry or company),
war stories (experience of a specific company
that implemented a leading practice), enablers (detailed analysis of tools that aid the
lead practices), lead-practice descriptions;
and reference materials (leading practices
source materials). This firm has broken its
leading practices knowledge base into eight
categories with the following approximate
number of entries
•
•
•
•
•
•
•
•
executive processes (1,163)
finance (1,488)
new business development (430)
knowledge management (222)
order management (702)
production and service delivery (421)
supply chain management (1,172)
support and shared service (945)
In addition, we can view their knowledge
base of leading practices from an industry
perspective (also with the number of entries):
• automotive (139)
• energy (715)
• financial services (622)
Understand
markets &
customers
Develop
vision &
strategy
•
•
•
•
•
•
•
health care (62)
insurance (96)
life science (50)
manufacturing (860)
retail (406)
service (47)
transportation (32)
Price Waterhouse’s Knowledge View. Price
Waterhouse was one of the first consulting
firms to develop a best-practices knowledge
base through its KnowledgeView system (see
Figure 3), which it generated solely for internal use and does not make directly available to
subscribers. KnowledgeView organizes processes into two basic categories: value chain
(productive) and support (overhead) processes. Reportedly, the system recently had over
4,000 entries. Price Waterhouse has translated
its ontology for KnowledgeView into five languages and uses it on five continents.
KnowledgeView’s knowledge base has
four views. The process view arranges the
knowledge base by different overall processes, such as customer service or design
and engineering. The industry view organizes
processes by different industries, such as
aerospace or chemical. The performance
measure view focuses on performance results
in terms of quality, cost, and time measures.
Finally, the best-practices database also has
an enabler view that captures information
about how the best practices are used in terms
of technology, people, other processes, and
organization structure.
Comparing best-practices models. The
models of the best-practices knowledge
Design
products
& services
Market
& sell
Produce &
deliver for
manufacturing
organization
Produce &
deliver for
service
organization
Invoice &
service
customers
Develop and manage human resources
Manage information
Manage financial and physical resources
Execute environmental management program
Manage external relationships
Manage improvement and change
Figure 2. Arthur Andersen’s Global Best Practices knowledge base.
IEEE INTELLIGENT SYSTEMS
.
bases for these three consulting firms are
similar. The value-chain process models
(Figures 2 and 3) from Arthur Andersen and
Price Waterhouse appear similar, with both
focusing on models that contain basic processes of marketing and sales, products and
services, production, and distribution and
customer service. Both models also include
similar views of the support functions, with
concern for human resources, financial resources, external relationships, environmental issues, information and systems, and business improvement.
In addition, the knowledge sources for
each of the best-practices knowledge bases
are also similar. Each provides best practices
by process, performance-measures benchmarks, industry-based processes information, and reference materials such as studies
and articles. Also, at least two—Ernst &
Young and Price Waterhouse—provide information about technology enablers, and those
same two are available only for internal
users, while KnowledgeSpace’s Global Best
Practices is available both internally and
externally to subscribers.
While there are some apparent differences,
these might result simply from differences in
disclosures about the best-practices databases. Ernst & Young indicates that it captures “war story” information. Arthur Andersen captures examples of companies applying
the practice, engagements by process, and
experts in each area.
In recognizing the high cost of developing
and maintaining these models and knowledge
Perform
marketing
and
sales
Define
products
and
services
bases, at least one consulting firm thought it
would be possible and desirable for firms to
form coalitions that could share development
costs. Other firms, however, apparently see
model and knowledge-base development as
providing a competitive advantage and so do
not want to share development.
How do KM ontologies work?
Ontologies are explicit specifications of
conceptualizations.7 These knowledge-based
specifications typically describe a taxonomy
of the tasks that define the knowledge. Within
the context of knowledge-management systems, ontologies are specifications of discourse in the form of a shared vocabulary.
They can differ by developer and industry.
Each consulting firm we have been examining has built or is building its own ontologies. Because these enterprise ontologies are
so costly to develop and maintain and are constantly changing, ontology or taxonomy issues
are emerging as some of the most important
problems in knowledge management.8
Need. A number of factors drive the need for
ontologies in knowledge management. First,
because knowledge-management systems
employ discussion groups, if users are going
to find discussion groups for either raising or
responding to an issue, they must be able to
isolate which groups are of interest. Ontologies serve to define the scope of these group
discussions, with ontology-formulated top-
Produce
products
and
services
Manage
logistics
and
distribution
Manage financials
Manage corporate services and facilities
Manage external relationships
Manage environmental concerns
Perform business improvement
Manage human resources
Provide legal services
Perform planning and management
Perform procurement
Develop & maintain systems and technology
Figure 3. Price Waterhouse’s KnowledgeView multi-industry process technology.
MAY/JUNE 1998
Perform
customer
service
ics used to distinguish between what different groups discuss. Without an ontology to
guide what the groups discuss, there can be
overlapping topics discussed in different
groups, making it difficult for users to find a
discussion that meets their needs. In another
scenario, multiple groups might address the
same needs, but possibly none would have
all the required resources to address a particular user’s needs.
Also, knowledge-management systems
must provide search capabilities. On the Internet, searches will often yield thousands of
possible solutions for search requests. This
might work in Internet environments, but it
generally is not appropriate in individual
organizational Intranet environments. To provide an appropriate level of precision, knowledge-management systems need to unambiguously determine what topics reside in
particular knowledge bases: “For the user,
the right [ontology] means the difference
between spending hours looking for information or going right to the source.”6
Furthermore, knowledge-management systems often provide filtering capabilities. Filtering systems (such as GrapeVine, http://
www.gvt.com) let one filtering source examine substantial amounts of information and
(hopefully) direct the information of interest
to the appropriate source. Those filters can be
either computer-based (such as intelligent
agents) or human-based. To use a filtering
system, users typically must specify keywords or concepts (depending on the nature
of the filtering system) that capture the nature of the desired knowledge. An ontology
therefore is essential to capture the set of
filter needs.
Ontologies facilitate reusability of artifacts
archived in the knowledge-management system as well. For example, consulting firms
typically archive an artifact referred to as a
proposal, which contains information about
proposals made to do revenue-generating
work for other firms. Proposal knowledge
bases can be quite large, so it is critical that
users be able to find what they are seeking.
To determine whether a previous proposal is
similar enough for potential reuse, proposal
users must have them categorized across the
common dimensions of an ontology, such as
domain or industry. For easier archiving,
ontologies need to categorize virtually all
artifacts in consultant knowledge bases.
Finally, knowledge-management systems
provide opportunities for collaboration and
use of expertise. However, without an appro37
.
priate set of ontologies, the lack of a common language might cause confusion in collaboration. Confusion could also arise in the
choice of collaboration partners. A knowledge-management system tries to facilitate
contact between experts and people in search
of their expertise. If a clear ontology of expertise is not available to support such contact, those expertise seekers will not find the
experts they are seeking or will find experts
whose services they do not need.
Desirable characteristics. When is one ontology seen as preferable to another ontology? What characteristics drive preferability? Knowledge managers have a number of
variables that facilitate choice.
• Cost-beneficial. For-profit businesses
such as consulting firms necessarily must
operate on cost-beneficial decisions. Because the generation of an ontology must
be cost-beneficial, firms might feel pressure to collaborate or use existing ontologies or derivatives of existing ontologies.
• Decomposable. For discussion groups and
knowledge bases to have a well-defined
audience, the ontologies on which they are
based need to be decomposable into relatively independent chunks of knowledge,
with little overlap. To provide for searchable knowledge, ontology chunks must
also be independent; otherwise, definitions of terms will overlap and searches
will be ineffective.
• Easily understandable. Clearly, an ontology’s intended users must be able to
understand and use it. Ensuring that materials are well-defined and graphically
illustrated will increase understandability. Generally, the easier something is to
understand, the less education will be
required for its use and the fewer mistakes
its users will make.
• Extensible. Knowledge changes, sometimes very rapidly, so ontologies must be
extensible to new concepts. Ontologies
developed for federations must also let
organizations add their own unique aspects to an ontology, so they must accommodate different organizational settings
and requirements.
• Maintainable. Knowledge is mobile, so
knowledge bases and ontologies need to
change over time. To facilitate that change,
knowledge must be packaged in a format
that is easily maintainable—as a database,
for example.
38
• Modular and interfaceable. Ultimately,
ontologies must interface with other
ontologies. Knowledge-management systems typically have multiple knowledge
bases and discussion groups, each potentially having its own ontology. There must
be a way to bring the multiple, potentially
conflicting ontologies together for search
and other capabilities.
• Theory/framework-based. If an ontology
is based on a theory, that framework can
facilitate many of the choices that need to
be made. A framework can mitigate issues
of redundancy and conflict. The frameworks in Figures 2 and 3 both can facilitate
categorization of best-practices knowledge.
• Tied to the information being analyzed.
While ontologies can exist within an
ONTOLOGIES PROVIDE THE
STRUCTURE TO FACILITATE
DRILLING DOWN IN THE
FRAMEWORKS TO PROVIDE
INCREASING LEVELS OF DETAIL
IN THE BEST-PRACTICES
KNOWLEDGE BASES.
organization, those organizations do not
operate independently of the rest of the
world. Any ontologies an organization
builds therefore must tie back to the rest
of the world. For example, if reengineering is a term used by the rest of the world,
that term should have a similar understanding within the firm. Otherwise, ambiguity can arise about definitions when
integrating external information. This
issue is particularly important if the
knowledge base is to contain information
from external sources, such as articles
from journals and magazines.
• Universally understood or translatable.
While an ontology ideally will be universally understood, in this era of multinational firms, that is at best an ideal. As
a result, if an ontology is not universally
understood, it should be translatable,
including from dialects.
Tools. A number of tools have been generated to attempt to facilitate development of
knowledge bases satisfying the criteria I have
just listed. These include database systems,
search tools, theory-based models, and visualization approaches.
• Databases and knowledge bases. Ontologies provide both a structure for developing knowledge bases and a basis for
generating views of knowledge bases
stored on a range of mediums, including
the Web, Lotus Notes, and general database systems such as Oracle. For example, Price Waterhouse uses a Lotus Notes
database to store Knowledge View, which
has different views on process, industry,
performance measure, and enabler. Most
importantly, the environment lets users
partially structure the knowledge and
facilitates the desirable characteristics
summarized above, such as extensibility
and maintainability.
• Search tools. Increasingly, as knowledge
bases grow larger, search is becoming a
critical aspect of knowledge-management
systems. A number of tools recently developed to facilitate search as a result:
Autonomy (http://www.agentware.com),
Magic Solutions (http://www.magicsolu
tions.com), and Open Text (http://www.
opentext.com), among others. Search
capabilities increase with an easily understandable ontology that ties to the rest of
the world’s existing information, without
conflicts.
• Theory/framework-based models. Developing an ontology requires the generation
of a dictionary that captures agreed-on
term meanings. That dictionary typically
is available as a knowledge base in the
knowledge-management system. Consulting firms base their ontologies on
frameworks, such as the models in Figures 2 and 3, that help them elicit dictionary items.
• Visualization tools. Visualization can be
one of the primary tools for managing
and using an ontology. Visualization lets
users view data from multiple perspectives so that they can identify data relationships in data that will help them find
what they are seeking. A visual version
of the ontology would allow a user to
visually follow a concept to its nearest
neighbors or analyze the overall space
for interesting related or unrelated concepts. A theory-based framework, such
as Figures 2 and 3, facilitates visualization of ontologies. In addition, knowlIEEE INTELLIGENT SYSTEMS
.
Coming this year in
IEEE Internet Computing
July
❖
August
Internet Search Technologies
Guest Editors Robert Filman, Lockheed Martin,
and Sangam Pant, Lycos
September
❖
October
Software Engineering
over the Internet
Guest Editors Frank Maurer, University of Calgary,
and Gail Kaiser, Columbia University
November
❖
IC Online
http://computer.org/
internet/
IEEE Internet Computing’s
companion webzine;
technology roadmaps to the
Internet for engineers and
computer scientists.
IC Online
Exclusives...
❖ Interviews: conversations with
December
the leaders in the field.
❖ Web resources: URLs,
Internet Security in
the Age of Mobile Code
Guest Editors Gary McGraw, Reliable Software Technologies,
and Edward Felten, Princeton University
event listings, forums.
❖ Industry reports:
news from the inside.
❖ And more all the time for
1998.
... you’ll only find
them here.
edge managers at consulting firms have
been examining a number of other tools
to generate a visualization of an ontology. In particular, knowledge managers
are studying how they can integrate tools
such as Perspecta (http://www.perspecta.
com) and InXight (http://www.inxight.
com) into their knowledge-management
systems for both users and ontology
management.
K
NOWLEDGE BASES AND ONTOLogies are closely related in knowledgemanagement systems.9 Ontologies provide
some structure for development of knowledge bases as well as a basis for generating
views of knowledge bases. For example, in
Price Waterhouse’s Knowledge View, the
knowledge base contains information on
process, industry, performance measure, and
enabler. That same information also serves
to provide four different views of the bestpractices knowledge base. Thus, each of
those views are representative of the structure of the underlying knowledge base, and
each are part of the best-practices ontology.
MAY/JUNE 1998
Also, at the highest level of abstraction,
ontologies define particular knowledge
bases—such as the best-practices knowledge
base. At lower levels, ontologies serve to
define models in particular knowledge
bases. Consulting firms find these models
quite useful: “Our experience taught us that
the common organizing framework was very
valuable—it provides us with a common and
understandable way to navigate through the
knowledge.”6 Ontologies provide the structure to facilitate drilling down in the frameworks to provide increasing levels of detail
in the best-practices knowledge bases.
References
1. J. Foley, “‘Giant’ Brains Are Thinking
Ahead—Professional Service Companies Are
Often Among the Early Adopters of New
Technology,” Information Week, No. 596,
Sept. 9, 1996.
2. Arthur Anderson, American Productivity &
Quality Center, Houston, 1997.
3. Knowledge, The Global Currency of the 21st
Century, Arthur Andersen, Chicago, 1997.
4. Leading Practices Knowledge Base, Ernst &
Young, Center for Business Knowledge,
Cleveland, 1997.
5. T. Davenport, “Some Principles of Knowledge Management,” http://knowman.bus.
utexas.edu/kmprin.htm, 1997.
6. Welcome to Knowledge View, Price Waterhouse, Dallas, 1995.
7. T. Gruber, “A Translational Approach to
Portable Ontologies,” Knowledge Acquisition,
Vol. 5, No. 2, 1993, pp. 199–220.
8. D. O’Leary, “Impediments in the Use of
Explicit Ontologies for KBS Development,”
Int’l J. Human Computer Studies, Vol. 46,
1997, pp. 327–337.
9. D. O’Leary and P. Watkins, “Integration of
Intelligent and Conventional Systems,” Int’l J.
Intelligent Systems in Accounting, Finance,
and Management, Vol. 1, No. 2, 1992, pp.
135–145.
Daniel E. O’Leary is a professor at the School of
Business of the University of Southern California.
He received his BS from Bowling Green State University, his masters from the University of Michigan, and his PhD from Case Western Reserve University. He is the Editor-in-Chief of IEEE
Intelligent Systems and is a member of the AAAI,
ACM, and IEEE. Contact him at the Univ. of
Southern California, 3660 Trousdale Parkway, Los
Angeles, CA 90089-1421; oleary@rcf.usc.edu.
39