COST-BENEFIT ANALYSIS FOR THE DESIGN OF PERSONAL
KNOWLEDGE MANAGEMENT SYSTEMS
Max Völkel and Andreas Abecker
FZI Forschungszentrum Informatik Karlsruhe, Haid-und-Neu-Straße 10-14, Karlsruhe, Germany
Keywords:
Knowledge management, personal knowledge, cost-benefit analysis.
Abstract:
Knowledge Management (KM) tools have become an established part of Enterprise Information Systems in
the recent years. While traditional KM initiatives typically address knowledge exchange within project teams,
communities of practice, within a whole enterprise, or even within the extended enterprise (customer knowl-
edge management, KM in the supply chain, . . . ), the relatively new area of Personal Knowledge Management
(PKM) investigates how knowledge workers can enhance their productivity by better encoding, accessing, and
reusing their personal knowledge. In this paper, we present a cost-benefit analysis of PKM where benefit
comes from efficiently finding task-specific, useful knowledge items, and costs come from search efforts as
well as externalisation and (re-)structuring efforts for the personal knowledge base.
1 INTRODUCTION
During the last decade, Knowledge Management
(KM) has reached a consolidated status as a manage-
ment discipline and has provided manifold impress-
ing business success stories in an economic world that
becomes increasingly knowledge-based. In balanced
socio-technical solutions, which holistically combine
organizational, psychological, and IT measures, En-
terprise Information Systems often play a central, en-
abling role. Originally KM functionalities enter the
area of standard EIS solutions (e. g. groupware func-
tionalities in Microsoft Sharepoint), and, vice versa,
sophisticated KM tool suites are stepwisely further
developed towards a central Information Backbone of
the enterprise (Maier, 2004).
However, in spite of all such developments, the
most important area of knowledge creation and pro-
cessing is still the most underdeveloped area of KM:
namely that of the individual knowledge workers
personal knowledge management (PKM) (Daven-
port, 2005). In the European Integrated Research
Project NEPOMUK
1
, we develop methods and tools
1
http:\\nepomuk.semanticdesktop.org Part of this
work has been funded by the European Commission in the
context of the IST NEPOMUK IP - The Social Semantic
Desktop, FP6-027705. Part of this work has been done in
WAVES – Wissensaustausch bei der verteilten Entwicklung
von Software, funded by BMBF, Germany.
for the Semantic Desktop, an integrative infrastruc-
ture for organizing, interlinking, querying and ex-
ploiting personal information from different everyday
applications through common semantics-based meta-
data. Later on, such personally created, collected,
and managed information shall partly be made
available to the peer group of the end user (their
team colleagues, project partners, buddy networks,
. . . ) through peer-to-peer technology, thus fostering
an integrated community-knowledge space.
As one of the first and most important, yet still
largely unsolved, questions arises that of appropriate
methods and tools to support (and seduce) the user to
articulate more, and more complex, parts of their ex-
pert knowledge in computer-based forms for stor-
ing, sharing, further developing it, etc. Obviously,
the more structure the user formally represents up-
front (e. g. by indexing or tagging content, by fill-
ing a given document template, by adding metadata,
etc.), the better prepared is such personal knowledge
for later efficient finding again, for context-specific
reuse, or also for sharing with other users. But this
is a central KM barrier: Chronically overloaded man-
agers, technicians, and creative workers simply do not
invest time and effort in extra work to prepare for
far-off, potential KM benefits. Hence, assuming that
knowledge workers act at least to a significant ex-
tent – rationally and economically, it is a central ques-
tion to better understand the trade-offs between actual
95
Völkel M. and Abecker A. (2008).
COST-BENEFIT ANALYSIS FOR THE DESIGN OF PERSONAL KNOWLEDGE MANAGEMENT SYSTEMS.
In Proceedings of the Tenth International Conference on Enterprise Information Systems - AIDSS, pages 95-105
DOI: 10.5220/0001713200950105
Copyright
c
SciTePress
costs and expected benefits of knowledge articulation
and of adding more structure to articulated knowl-
edge. Insights in this area are required for building
suitable Personal KM systems which will be accepted
and used by knowledge workers and which can prove
their Return-on-Investment in an enterprise setting.
Outline. We first describe the relations between
knowledge management (KM), personal information
management (PIM), and personal knowledge man-
agement (PKM). In brief, PKM can be seen as the
KM perspective on PIM or the personal perspective
on KM. Then we look from a high-level on costs and
benefits in PKM. We develop a unified knowledge
model (UKM) and explain how it can represent doc-
uments and ontologies in a unified way. Using the
UKM, we refine the cost model and explain general
cost drivers and benefits of PKM. Finally we apply
the resulting formula as an example to the possible
extend on a semantic wiki.
1.1 Knowledge Management
Drucker (1985) was among the first to use the term
knowledge worker when stating
The most important contribution of manage-
ment in the 20th century was to increase man-
ual worker productivity fifty-fold. The most
important contribution of management in the
21st century will be to increase knowledge
worker productivity hopefully by the same
percentage. [...] The methods, however, are
totally different from those that increased the
productivity of manual workers.
At times when knowledge management was be-
coming popular, Nonaka and Takeuchi (1995) pub-
lished a book on knowledge processes which de-
scribes two basic states of knowledge: tacit (implicit)
and external. Later works (Despres and Chauvel,
2000) conclude that external and internal knowledge
are two extremes on a spectrum, but do not exist in
reality. Maurer (1999) states that knowledge resides
in the heads of people and the computer can only
store "computerized knowledge" which is to be un-
derstood as “shadow knowledge”, a “weakish image”
of the real knowledge. The high-level processes in
knowledge management are externalisation, internal-
isation, combination and socialisation (Nonaka and
Takeuchi, 1995). North (2007) defines knowledge
work as “work based on knowledge with an immate-
rial result; value creation is based on processing, gen-
erating and communicating knowledge.
2
2
translation by the author
1.2 Personal Knowledge Management
The knowledge-based organisation is no more effec-
tive than the sum of its knowledge workers’ effec-
tiveness (Davenport, 2005). Knowledge can be em-
brained (conceptual, implicit), embodied (tacit, im-
plicit), encultured (shared beliefs), embedded (in pro-
cesses) or encoded (symbolic, external) (Blackler,
1995).
We use the term personal knowledge management
(PKM) to denote the process of the individual to man-
age knowledge. In contrast to general knowledge
management, PKM denotes the perspective of the in-
dividual. This perspective is potentially better suited
to explain individual motivations and behaviour, even
in organisational contexts.
The term personal knowledge has already been
used in (Polanyi, 1958), the term PKM appears
in (Frand and Hixon, 1999; Mitchell, 2005). The
fields “ePortfolio” and “personal learning environ-
ment” deal with similar topics.
Higgison (2005) defines personal knowledge
management as “managing and supporting personal
knowledge and information so that it is accessible,
meaningful and valuable to the individual; maintain-
ing networks, contacts and communities; making life
easier and more enjoyable; and exploiting personal
capital”.
Organizing information is a central part of the
inquiry process focused on making the con-
nections necessary to link pieces of informa-
tion. Techniques for organizing information
help the inquirer to overcome some of the lim-
itations of the human information processing
system. In some ways the key challenge in
organizing information is for the inquirer to
make the information his or her own through
the use of ordering and connecting principles
that relate new information to old information.
. . . (Avery et al., 2001)
The related topic of Personal information manage-
ment (PIM) got academic attention around 2005
when the first PIM workshop was held. The report
of the second workshop (Jones and Bruce, 2005) de-
fines the term Personal Space Of Information (PSI)
as the space that includes all the information items
that are, at least nominally, under that person’s control
(but not necessarily exclusively so). This includes a
number of analog and digital storage locations.
There is no sharp distinction between PIM and
PKM, rather a difference of scope and perspective.
PIM focuses on managing all the information around
an individual, i. e. only encoded knowledge. PKM
deals with embrained, embodied and encoded knowl-
ICEIS 2008 - International Conference on Enterprise Information Systems
96
edge, i. e. mostly with personal, self-authored arte-
facts. This paper tackles the continuum between em-
brained and encoded knowledge, with regards to costs
and benefits.
Jones et al. (2001) introduces the problem of
“keeping found things found” which reports on the
difference between knowing something and merely
storing information. E. g. obviously an unread email
can be managed in the sense of information manage-
ment, but the knowledge one can get from the email
has not yet been realised.
The basic processes in PIM are (Jones and Bruce,
2005): 1. keeping (input of information into a PSI);
2. finding/re-finding (output of information from a
PSI); and 3. meta-activities like mapping between in-
formation and need, maintenance and organisation .
The role of handling an external memory is apparent.
In the next section we analyse costs associated to each
phase. Note that the basic processes are the same for
PIM and PKM, the difference is the perspective. PKM
emphasises information from the users mind encoded
by the user himself. PIM has a broader focus. PKM
focuses on the knowledge in the users mind evoked
by a piece of information presented by the system.
Many real-world notes seem to fall between em-
brained and encoded (Blackler, 1995) or between im-
plicit and external (Nonaka and Takeuchi, 1995). E. g.
a personal note with the content “milk” is not really
an external, encoded representation of the knowledge
“I need to buy some milk today”. On the other hand,
without the note, the action to buy milk might not hap-
pen. Different audiences and topics require different
degrees of explicitness (Boettger, 2005).
Definition. A knowledge cue is an artifact which
evokes (acts as a cue to) some kind of knowledge
in the readers mind. I. e. knowledge is either re-
activated by consuming the artifact representing the
knowledge cue or new knowledge is obtained by con-
suming the information, e. g. when reading a book.
The reader and author of a knowledge cue are the
same person; other persons might not be able to gain
knowledge from using the artifact representing the
knowledge cue.
It is not possible to define precisely the amount
of knowledge contained in one knowledge cue.
Roughly, one knowledge cue evokes one concept or
family of related concepts. As an example, a shop-
ping list consisting of n items contains n knowledge
cues. A description of a single business innovation
could be treated as one cue. Bernstein et al. (2007)
uses the term “information scrap” to denote one or
several knowledge cues. On aspect of PKM is effi-
cient and effective management of knowledge cues.
Use
Benefit of use (B)
Creation
Cost of Creation (C
C
)
Figure 1: A Simplified Model for Cost/Benefit-Analysis in
PKM without external memory stores.
2 COSTS AND BENEFITS IN PKM
We need to think in terms of investment, allocation of
costs and benefits between the organizer and retriever
(Glushko, 2006)[p. 24-31]. In PKM, organizer and
retriever are the same person. Different from an or-
ganisational context, there is a personal motivation to
organize knowledge.
Jones and Bruce (2005) reports on difficulties of
evaluating PIM systems, as almost no person is will-
ing to try out an experimental, potential unstable PIM
system for a long period of time (months or years) for
critical personal tasks such as remembering appoint-
ments, managing email or keeping personal notes.
The analysis in this paper can help to evaluate sys-
tems used for PKM, e. g. also PIM systems.
This section analyses costs and benefits of an in-
dividual doing PKM from the perspective of a neutral
observer. First we describe the PKM process without
external storage and then extend this model.
2.1 Knowledge Creation and Use
If no external tools are used, the PKM process
(c. f. Fig. 1) consists of
Knowledge creation at a cost C
C
and
Knowledge use with a benefit B.
The cost of knowledge creation C
C
is the amount of
time, money and effort an individual spends on think-
ing, researching, experimenting, and learning. Cost
can maybe best be measured by the amount of time
spend but how to measure the value? Maybe the
time needed to re-create the knowledge? But if it
takes long to re-create some knowledge that has a low
benefit, then one would intuitively not assign a high
value to it. Also knowledge might be cheap to create
and store now but much more costly to re-create later,
e.g. Where have I been on a given day in 1999?
The value of knowledge does not exist as such
(Iske and Boekhoff, 2002); it depends highly on the
COST-BENEFIT ANALYSIS FOR THE DESIGN OF PERSONAL KNOWLEDGE MANAGEMENT SYSTEMS
97
Use
Later …
Personal
Information
Space
Externalisation
(authoring,
structuring)
Personal
Information
Space
Retrieval
Cost of Externalisation (C
E
) Cost of Retrieval (C
R
) Benefit of use (B)
Creation
Cost of Creation (C
C
)
Figure 2: A Simplified Model for Cost/Benefit-Analysis in PKM. Goal: more benefit than costs (B > C
E
+C
R
).
task. The value of some knowledge can be defined
as “increment in expected utility resulting from an
improved choice” made possible by this knowledge
(Varian, 1999). I. e. one estimates the value of the
state of the world that would result from actions per-
formed in absence of the knowledge (V
1
) and com-
pares it with the value of the state of the world result-
ing from actions taking in presence of the knowledge
(V
2
). Then the benefit B of having this knowledge for
this task is the difference in value, i. e. B = V
2
V
1
. B
then represents the additional created value or saved
costs. B can (in theory) be measured in money, saved
time, improved quality or better emotions. In practice,
measuring the value of the state of the world is often
hard to quantify, i. e. consider long-term effects and
the difficulty to measure quality of better emotions.
Most approaches resort to measure only costs (Feld-
man et al., 2005). Economists use an abstract “utility
value” without defining a unit of measurement.
For comparing different PKM systems one needs
a pragmatic way to take the benefit into account. We
believe in PKM, the value of a certain piece of knowl-
edge with respect to a task can be estimated by indi-
viduals on a Likert scale (Likert, 1932). The added
value of using a knowledge management system for
a given period of time (t) is the overall cost-benefit
gain G which can be approximated by summing up
all benefits B and subtracting the sum of all costs C
(G = B C). For a given amount of knowledge one
can compare the costs of different PKM systems.
2.2 Using External Storage
In this section we extend the simple model introduced
in Sec. 2.1 with an external storage system. The pro-
cess of managing knowledge cues can be represented
as (c. f. Fig. 2):
1. Creation: Knowledge is created at some costs C
C
.
2. Externalisation: Some parts of the implicit knowl-
edge become external. Knowledge cues are cre-
ated. These user has externalisation costs C
E
.
3. Time passes by and the author might start to for-
get some or all details of the articulated knowl-
edge. Sometimes even the knowledge to know
something is forgotten as well.
4. Retrieval: At a certain moment, while perform-
ing a certain task, the user initiates a retrieval pro-
cess in his PKM system. As information retrieval
system have become faster, the classic informa-
tion retrieval measure of “time to execute query”
becomes less relevant to determine the costs per-
ceived by the user. The human-computer interac-
tion becomes more often the bottleneck and cost-
driver of efficient knowledge work.
After having executed a query or performed a
browsing step, the user reads the search results,
and refines the search query. After some steps, the
user either found one or several matching knowl-
edge cues or cancels the search with no result. The
process of reading through a list of search results
takes time and therefore adds to the search costs.
If a knowledge artifact is long in size, the time to
read through it takes longer. If the desired knowl-
edge is only a part of the artifact, reading through
the artifact is thus additional search cost. All these
costs are subsumed under retrieval costs C
R
.
5. Usage: If results where found, the user has some
benefit from having available the external knowl-
edge or from remembering knowledge from the
knowledge cue.
Esser (1998) analyses factors that determine when
and which external memory humans use. Three vari-
ables were observed: Expected likelihood of success-
ful remembering a piece of knowledge when stored
in an external store, cost of storing it there and most
importantly: perceived value of the knowledge to be
stored. The higher the perceived importance of re-
membering the knowledge, the higher costs for stor-
age were accepted.
The basic hopes of each person doing PKM should
be:
ICEIS 2008 - International Conference on Enterprise Information Systems
98
C
E
C
R
Zone of lo
w
est
total
costs (C
E
+ C
R
)
Figure 3: Assumed relation between externalisation costs
(C
E
) and retrieval costs (C
R
).
There is more benefit of having the knowledge
(again) than it had cost to manage it, i. e. C
E
+
C
R
< B
Managing knowledge externally is cheaper than
re-creating it from scratch, i. e. C
E
+C
R
< C
C
.
Estimating the benefit of storing an item for later use
(C
E
+ C
R
)) compared with the expected costs to re-
generate the contained knowledge later (C
C
) is cer-
tainly hard. An assumption underlying this paper is
that better organisation, structuring, and formalisation
of content is expected to lower the costs of retrieval
(Glushko, 2006).
2.3 Knowledge Representation
For a long time in history, knowledge cues were only
stored in documents, first analog then digital. Digital
document retrieval is the prime application domain of
information retrieval (IR) techniques. The field of IR
has a history of quantitative research, mostly focusing
on precision and recall. These measures were first de-
fined in Cleverdon et al. (1966). However, Cleverdon
et al. (1966) proposed to measure not only these two
factors but also “the extent to which the system in-
cludes relevant matter”, and “the effort involved on
the part of the user in obtaining answers to his search
requests”. Maybe because these two factors cannot so
easily be measured automatically, they were mostly
ignored by IR research.
The TREC conferences have heavily influenced
IR research. In the last ten years it ran the novelty
track, in which each sentence of a document was re-
garded as an information item on its own. The ques-
tion answering track goes in a similar direction, here
the user is not querying for a set of documents but for
a concise, factual answer to his question. Both tracks
show a tendency towards smaller content granularity.
A fact given within a document requires on average to
read half of the document until the fact is found. The
granularity of information thus influences its retrieval
costs. The coarse granularity of documents leads to
high costs for locating relevant parts of documents,
e. g. for re-use, aggregate queries, or question answer-
ing. The extreme case of a document is a mere list of
sentences, devoid any further structure.
Data bases and ontologies allow for more struc-
tured access, i. e. browsing and searching. In contrast
to documents, data bases and ontologies allow to re-
trieve sets of items together with relevant properties.
Ontologies (and some database systems as well) al-
low to answer queries to which the answer has only
implicitly been entered. It requires effort to structure
and formalise knowledge to make it fit into a database
or ontology, but retrieval abilities are also higher
ontologies are built for re-using knowledge.
Taken to its extreme, one could try to formalise all
personal knowledge, leading to exorbitant externali-
sation costs and very low retrieval costs. In reality,
one can expect a “sweet cost spot” for the total costs
(C = C
E
+ C
R
) as depicted in Fig. 3. For efficient
knowledge work, a user should be able to work close
to the sweet spot, using something half document and
half ontology.
3 UNIFIED KNOWLEDGE
MODEL (UKM)
In this section we present a unified knowledge model
(UKM). It is unified in the sense that it can repre-
sent a range of existing knowledge formalisms such
as documents and formal ontologies. The UKM is
also unified in the sense that it represents both textual
content and relations among content items. These re-
lations can be formal or informal. The term knowl-
edge model is used instead of ontology to emphasize
a large amount of stored textual content.
There is a basic tradeoff between acquirability of
a knowledge representation language and it expres-
sive power (Gruber, 1989). The purpose of the unified
knowledge model (UKM) is to analyse costs in PKM
processes, not to be used by end-users for their PKM.
A slightly modified version of the UKM is used for
PKM. This version is called Semantic Web Content
Model (SWCM) (Völkel, 2007). The main difference
between SWCM and UKM is that SWCM allows dif-
ferent sizes of items whereas UKM has a maximal
item content size of one sentence. SWCM has also
more features which make it more usable for PKM.
To be able to model both documents, semantic
nets and the continuum between these two models,
the UKM must take into account different “degrees of
formality”, a notion introduced by Lethbridge (1991)
3
Granularity is an important cost driver (c. f.
Sec. 2.2). The smallest units of content that make
3
Lethbridge uses a kind of semantic net with concept
and relation subsumption hierarchies.
COST-BENEFIT ANALYSIS FOR THE DESIGN OF PERSONAL KNOWLEDGE MANAGEMENT SYSTEMS
99
sense to a human are single or multiple words (encod-
ing concepts) or sentences (encoding facts or ques-
tions). The UKM must also be able to represent for-
mal statements. Given these two constraints, we de-
fine:
Definition. A knowledge item is the smallest unit
of content in the UKM. A knowledge item is either
a snippet of content which can contain something
between a single word up to a sentence, or
a knowledge item is a statement between other
knowledge items. A formal statement is mod-
elled similar to a semantic net or the RDF stan-
dard (Klyne and Carroll, 2004) as a triple of
three items, the middle item playing the role of
the relationship type. Representing statements
as knowledge items themselves allows full meta-
modelling.
A knowledge item is a technical counterpart of a
knowledge cue. Knowledge cues can also be persisted
in other forms, e. g. in real-world objects such as a
knot in the handkerchief.
For cost analysis we need to formalise the UKM.
A knowledge model K is defined as a set of nodes N
and arcs A, i. e. K := {N, A,C, f , S}, where the arcs
are defined as A := {N × N × N}. We use items as
relations to allow the user to extend the vocabulary
and to represent arbitrary knowledge. A function f
can assign each node a piece of content from the set
C of all content: f : N 7→ C. A content snippet c is
a simple linear stream of symbols (S) devoid further
machine processable structure. We ignore aspects of
natural language processing as we care about measur-
ing explicitly structured knowledge. All structural as-
pects of a document or other representation format are
expressed as relations between content items c C.
Modelling the symbols explicitly allows to formalise
the ability of the computer to map search queries via
bag-of-word, vector space, or other IR-models to the
content.
In the next two subsections we analyse how UKM
can represent existing knowledge representation for-
mats.
3.1 Documents
Most importantly, we analyse documents, which have
been used for several thousand years now. How to
model the structure and content of documents?
A French team of over 50 researchers analysed
the term document in depth (Pédauque, 2003) and
gives three co-existing definitions of the term "doc-
ument": (i) Document as form, where a document
is seen mostly as a container, which assembles and
structures the content to make it easier for the reader
to understand it. (ii) Document as sign, which em-
phasizes the argumentative structure of the content.
Also, a document that can be referenced acts as a sign
for its meaning. (iii) Document as medium, concen-
trates on the "reading contract", that is the intention
or assumption of the author what will happen with the
document.
A document contains a number of knowledge
items (c. f. Sec. 3). This act of “packaging” together
a set of knowledge items influences the interpretation
of each item by the reader. A document is a knowl-
edge artefact consisting of several layers. Aspects of
information in a document are:
Reference-ability. Once a document is published,
the reference can act as a placeholder for the con-
tent expressed within. A reference to a document
can act as a meta-symbol on top of the knowl-
edge items the document contains. The usage of
document references as symbols allows a docu-
ment to “participate” in conversations, which lead
to scholastic methods and modern academia. To
represent a document in UKM, we use one knowl-
edge item to represent the root of the document
and store the title as the content of it.
Process Metadata. Each document is written by a
number of authors for a certain audience with a
certain goal. By sending this process metadata
along with the document the reader has the ability
to put the document in context and interpret it bet-
ter. Such metadata is used by the reader as a frame
of reference for interpretation and for search. In
UKM it is modelled as additional items linked to
the document root, similar to the way how RDF is
used.
Linearity. A document can typically be read from
start to end by navigating through all contained
knowledge items. This is modelled via an hasNext
relation between the items holding the document
sentences.
Visual Structure. A document is not only a stream
of sentences, but uses type-setting, i.e. bold, ital-
ics, different font styles and size, and placement
of figures. For sake of simplicity, we ignore these
properties in the UKM.
Logical Structure. The visual structure is used to
encode a logical structure consisting of i.e. para-
graphs, headlines, footnotes, citations, and title.
The logical structure makes it possible to refer-
ence smaller, meaningful parts within a document,
i.e. "Sec. 4.2". Following the approach of Groza
et al. (2007), we model a document structurally as
a tree consisting of root, sections nested into each
ICEIS 2008 - International Conference on Enterprise Information Systems
100
other, paragraph and sentence. Sections can also
contains figures and tables, which are not further
modularised. In our cost model we introduce a re-
lation hasPart which is used to model the different
kinds of containment. To distinguish the different
types of structural unit we use a relation hasType
and a number of type-items, e. g. section, para-
graph, etc.
Argumentative Structure. On top of the linear con-
tent, a document follows an argumentative struc-
ture to convey its content to the reader. Argu-
mentative structures appear on all scales. A typ-
ical structure is the "Introduction - Related work
- Contribution - Conclusion"-pattern of scientific
articles. On smaller scales, patterns like "claim-
proof" and "question-answer" are used. Groza
et al. (2007) also describes ways to encode argu-
mentative structures.
Content Semantics. Documents content’s mean
something. Building upon logical and argumen-
tative structure, the author encodes statements
about a domain within the content. We allow to
store semantic statements in the UKM.
3.2 Ontologies
How to encode ontologies in UKM? A mapping from
RDF to UKM is pretty straightforward. Each triple in
RDF consists of URIs (U), blank nodes (B) and lit-
erals (L) and is of the form (U, B) × (U) × (U, B, L).
First we replace all literals with nodes and assign the
literals content as node content, f (n) =literal. Next
we replace all URIs and blank nodes with nodes, us-
ing the same nodes where the same URI or same blank
nodes is denoted. Now each triple is converted to the
form N × N × N and can be stored in UKM. Subtleties
such as language tags and data-types of literals can be
stored as further statements in UKM, so there is no
information loss.
We expect future PKM systems to allow mod-
elling textual and semantic content in the same en-
vironment, as described in (Bettoni et al., 1998; Lud-
wig, 2005; Oren et al., 2006)
4 A COST-MODEL FOR PKM
In this section we use the UKM to measure the ex-
ternalisation and retrieval costs in PKM system. We
have the following basic factors for costs and benefits:
Each knowledge cue x that is externalised has ex-
ternalisation costs C
E
(x). We detail these costs in
the next section.
For each task t T, the user has the option to
search for knowledge cues. This has retrieval
costs C
R
(t). Note that a user might retrieve an
item several times or not at all.
Retrieved items have a benefit B(t) for the given
task t.
The overall process has thus the following gain:
G =
t
B(t) (
x
C
E
(x) +
t
C
R
(t))
=
t
(B(t) C
R
(t))
x
C
E
(x)
4.1 Externalisation Costs
C
E
can be divided into cost of authoring the content
(C
A
) and costs of (re-)structuring existing knowledge,
classifying new or existing items or linking between
items (C
S
). Linking items can also be an act of for-
malisation if the relation is specified with a relation
that has a formal semantics. Hence C
E
= C
A
+C
S
.
Let N be the set of all knowledge cues in the sys-
tem. Cost of content externalisation is correlated to
the size of externalised artefacts. E. g. writing more
words takes more time. Let | n
j
| be the size of the
jth item, measured in the number of symbols it con-
tains (c. f. UKM). Articulating a single symbol costs
c
s
. Articulating the jth item costs | n
j
| c
s
. Then
C
A
=
j
|n
j
|c
s
.
The structuring costs C
S
will often involve more
then one item, e. g. when linking two items. Structur-
ing is the process of linking, tagging, typing and cate-
gorising items. The cost of restructuring are indepen-
dent of the items size. The more items a knowledge
base contains, the more effort it might take to find
the right element to link another item to. Structuring
is expected to make more of the knowledge accessi-
ble to the computer, which should enable to answer
queries with better (more) results. Note: The structure
of a knowledge models contains itself knowledge. It
is not possible to specify the structuring costs per se,
but we can expect the degree of formality of a knowl-
edge model to correlate with the structuring costs
spend. The degree of formality d
f
can be measured
by the amount of formal statements (|A|)compared to
the number of knowledge items, similar to the defini-
tions given in (Lethbridge, 1998) as d
f
=
|A|
|N|
. If we
further assume a fixed cost c
f
for articulating a for-
mal statement, we can estimate C
S
= |A|c
f
. We get
the total externalisation costs
C
E
= C
A
+C
S
=
j
|n
j
|c
s
+ |A|c
f
(1)
This equation assumes that no content and no for-
mal statement is ever deleted or changed. But in re-
ality, the cues need to be maintained to keep or im-
prove their value over time. E. g. some knowledge is
COST-BENEFIT ANALYSIS FOR THE DESIGN OF PERSONAL KNOWLEDGE MANAGEMENT SYSTEMS
101
no longer applicable or needs to be updated to reflect
changes in the world. Informal ideas undergo sev-
eral transitions until some of them might become text
books (Maier and Schmidt, 2007).
Some structuring operations could as a side ef-
fect increase (split) or decrease (merge) the number
of cues. We ignore changes to the number of cues
for sake of simplicity. Deleting outdated or erro-
neous knowledge could improve the value of using
the knowledge model quite a lot, but some costs do
occur for these maintenance tasks, too. Therefore in-
stead of measuring the knowledge model as such, we
measure the costs of the operations that lead to the
current state, i. e. all operations performed.
Let c be a function that assigns each operation
some costs. Basic operations on a model are:
add content (content
a
) Adding m symbols to a
knowledge cue costs m × c(c
sa
) with c
sa
being the
costs of adding one symbol.
delete content (content
d
) Deleting m symbols from
a knowledge cue costs m × c
ds
with c
ds
being the
cost of deleting a symbol. Deleting has often
lower cost than adding, e. g. when deleting a com-
plete item in the user interface which causes dele-
tion of many symbols.
Cost of updating can be modelled as the sum of
deletion costs and addition costs.
add formal statement (stmt
a
) Adding a formal
statement. The cognitive costs (measured in time
and ultimately money) should vary according
to the severity of the formal statement. E. g. it
should take less time to create a hasPart or hasIn-
stance relation than a hasSubclass statement.
These differences will be taken into account in
future versions of the cost model.
delete formal statement (stmt
d
) Deleting a single
statement could have dramatic effects, depending
on the used inference rules.
We take the restructuring operations into account and
define n(content
a
) as the total number of added sym-
bols and respectively n(content
d
) as the number of
deleted symbols. Let τ be the total costs of an opera-
tion, calculated by multiplying the number the opera-
tion is performed with the costs of the operation, i. e.
τ
op
= n
op
c
op
. We can reformulate the externalisation
costs as
C
E
= τ
content
a
+ τ
content
d
+ τ
stmt
a
+ τ
stmt
d
4.2 Retrieval Costs
In order to precise the relation of structures in the
knowledge base and search costs one first needs to
develop a unified model for the search process, which
does not exist yet. A first work in this direction is
the “information foraging process” (Pirolli and Card,
1995). There are three basic ways to retrieve infor-
mation when interacting with an information system
(Bates, 2002):
Browsing a collection of items related to the infor-
mation need. In principle, two kinds of collec-
tions are possible: explicit, i. e. created by a user,
and implicit, i. e. the members in the set are deter-
mined by a (semantic) query. Toms (2000) and
Teevan et al. (2004) emphasise the importance
of finding information “by accident”, e. g. when
searching for something else. Such (re-)findings
are important for creative processes and knowl-
edge creation.
Formally, browsing is the act of scanning a list of
items and evaluating each of them for relevance.
Evaluating a single item has the costs e. The user
is free to stop evaluating items from the list at any
time.
Searching denotes the process of executing a query
(e. g. keywords) and refining it until the top re-
sults are relevant to the information need. Seman-
tic queries, utilising knowledge indirectly for in-
ferencing, also fit into this category.
Formulating a query has the costs q. Systems that
allow several kinds of queries need different val-
ues for q to model the difference in cost. After
each search-step the user is confronted with a list
of search results which need to be evaluated, sim-
ilar to browsing.
The search costs depend on the complexity of the
query and the structure of the knowledge base. A
complex query has a higher cost to be formulated,
but has the ability to return exactly the required
cue. Simpler queries return usually too many re-
sults and need refinement. From a users perspec-
tive, starting with simpler queries that are gradu-
ally refined is more economic than asking directly
a complex query. The interactive refinement pro-
cess gives earlier feedback about how many re-
sults are returned, which guides query refinement
until the query is complex enough to filter out the
desired cues. This way, queries do not become
more complex than needed.
Following links should not be confused with brows-
ing. A common practice in large search spaces for
which neither suitable collections nor query terms
are known is to explore e. g. citation links. Fol-
lowing links is thus a kind of associative retrieval.
Following a link has the costs l.
ICEIS 2008 - International Conference on Enterprise Information Systems
102
A complete search process involves typically all three
kinds of operations. Instead of measuring each step,
we model the complete retrieval process as a process
that involves some costs C
R
(t). These costs can be
broken down to C
ql
(t) for formulating queries and fol-
lowing links and costs for evaluating items.
Assuming the user evaluates k(t) items in the re-
trieval process, we can define task-specific precision
p
t
and recall r
t
(Van Rijsbergen, 1979). As a refine-
ment of this idea, an relevant item has a certain bene-
fit for the given task (in range 0 . . . 1). We define the
benefit of an item j for a given task t to be v
j
(t). For
irrelevant items, v
i
(t) is zero.
Let k(t) be the total number of all retrieved items
in the search process. There is a certain cost e to
evaluate each item in order to be able to decide if the
knowledge represented in the item is relevant for the
task. We assume the effort of evaluating a single item
is not correlated to precision and recall of the com-
plete process. The complete costs of the retrieval pro-
cess for task t are then C
R
(t) = C
ql
(t) + k(t)e.
Only retrieved knowledge cues can bring benefit
for the user. Knowledge that is never used is of zero
value. The complete benefit B(t) of the k(t) retrieved
items is B(t) =
j
v
j
(t). Both p
t
or k(t) might also
be zero. Assuming all retrieved items in a task are
either relevant (value = 1) or not, we can simplify the
formula as B(t) = p
t
k(t).
For each retrieval process we get:
B(t) C
R
(t)
= p
t
k(t) C
ql
(t) + k(t)e
= k(t)(p
t
e)C
ql
(t)
Thus the query formulation costs are a kind of fixed
costs, whereas the relation between precision and
evaluation costs decides if the whole retrieval process
was worth the hassle. Interestingly, higher recall val-
ues seem in the light of cost-benefit analysis less rele-
vant than high precision values. Precision in retrieval
values is typically heavily dependent on the degree of
structuredness and formality of the data.
If we analyse the equation, we see three factors
than can negatively influence the gain of using a PKM
system:
1. If the user does not try to retrieve knowledge for a
task t;
2. if no or too few relevant items are retrieved, i. e. if
the precision too low;
3. if the value of the successfully retrieved items is
too low, i. e. results fit, but of too low value.
Factor (1) is addressed by (Cutrell et al., 2006) which
proposes to automatically start a search when certain
triggers are encountered. Factor (2) is addressed by
works in knowledge articulation and modelling, in-
formation retrieval and improved search algorithms.
Factor (3) can maybe only be addressed by personal
experience or training.
4.3 The Complete Cost Function
Stitching the parts together we get:
G =
t
(B(t) C
R
(t))
x
C
E
(x)
=
t
k(t)(p
t
e)C
ql
(t)
τ
content
a
τ
content
d
τ
stmt
a
τ
stmt
d
As we see one of the most important cost drivers is
the question how well the costs spend on externalisa-
tion and query formulation can improve the precision
of the retrieved items. Note that most works address
improvements of precision only by looking at the data
as “given”. In PKM, this is not true, as the knowledge
items are authored and retrieved by the same audi-
ence.
5 APPLYING THE COST MODEL
TO SEMANTIC WIKIS
As an example, we apply the cost model (i. e. Sec.
4.1) to a semantic wiki, in this case to Semantic Me-
diaWiki (SMW) (Krötzsch et al., 2006). First we need
to represent the data model of SMW in terms of the
UKM. Each wiki page in SMW can be regarded as a
knowledge item. SMW has two types of formal state-
ments: Type (a) links a page to another page; type
(b) links a page to a value stored within that page.
We model type (a) as a formal statement in the UKM.
Type (b) links can be represented as a link from the
page to a knowledge item which contains that data
value. For structuring, SMW allows to create seman-
tic links or put wiki pages into categories. Categories
can be modelled as knowledge items containing only
their name and being linked to each category member.
Addition and deletion of category links and semantic
links can be measured as stmt
a
and stmt
d
.
Note: As SMW uses MediaWikis versioning his-
tory, one could in theory calculate all modelling op-
erations that ever happened. The time to externalise
knowledge as text or wiki links could be measured. In
the future, we consider to perform this kind of eval-
uation on existing public instances of SMW. Via ad-
ditional usage logs one could determine the average
time e. g. it takes to pose a query.
Although there has been until today no study on
personal wiki use, many of our colleagues do use se-
mantic wikis and in particular SMW as their PKM
tool. SMW allows all three basic kinds of retrieval:
COST-BENEFIT ANALYSIS FOR THE DESIGN OF PERSONAL KNOWLEDGE MANAGEMENT SYSTEMS
103
browsing, searching and following links. A user can
browse e. g. all pages in a category, or all members of
a list generated by an a semantic query embedded in a
page. For search the user can either perform standard
keyword search or pose semantic queries to the sys-
tem. The ability to follow links is obvious in a wiki.
These properties make SMW an ideal study ob-
ject for PKM, as it uses almost all imaginable ways to
state and retrieve knowledge – only statements about
statements are not possible in SMW.
6 RELATED WORK
There has not been much work on estimating cost and
benefits in PKM.
A related work done by Bontas et al. (2006) in
the area of ontology engineering does not fit our use
case as the use of a personal knowledge model is not
a linear, planned process with the goal of creating a
formal representation. Rather the contrary is true: An
individual is always reluctant to formalise anything,
because its unclear if the extra effort will ever pay off.
(Lethbridge, 1998) shows metrics for concept-
oriented knowledge bases, but does not take costs into
account.
7 CONCLUSIONS AND
OUTLOOK
This paper makes a first attempt to understand the
complete PKM process in order to help design bet-
ter PKM tools. The overall benefit of using a PKM
system could be characterised by summarizing over
the successfully retrieved knowledge items (content
or formal statements) for each task. Costs could be
characterised as the sum of the costs of all authoring
and structuring efforts. A quantification of the effect
more structuring has on lower retrieval costs (or im-
proved benefit) cannot be stated unless the semantics
of the formal statements and details of the search pro-
cess (browse, search, follow links) are specified. Thus
the resulting formulas can serve only as a conceptual
framework or starting point for tool-specific measure-
ments.
7.1 Future Work
As future work we intend to develop automatic mea-
sures of the information content of a knowledge
model, by counting the size and used symbols (here:
words) of each knowledge item as well as the num-
ber and kind of semantic links. We have to take the
semantics of the knowledge model into account, as
some formal statements have a much higher influ-
ence than others, c. f. ontology evaluation (Vrandecic,
2006). SMW offers only a transitive category hierar-
chy, hence the transitive closure can be calculated and
taken into account. Counting the number and kind
of modelling steps used in the history of a semantic
wiki is also planned. To estimate value v
i
, precision
p
i
, search costs c
i
and number of returned items k
i
in
search processes, we intend to perform a diary study.
Another important aspect of future research is an
investigation in which way investments in structuring
(C
S
) can lower the cost of retrieval (C
R
), e. g. by im-
proving p
i
or k
i
.
REFERENCES
Avery, S., Brooks, R., Brown, J., Dorsey, P., and O’Conner,
M. (2001). Personal knowledge management: Frame-
work for integration and partnerships. In Proc. of AS-
CUE Conf.
Bates, M. (2002). Speculations on browsing, directed
searching, and linking in relation to the bradford distribu-
tion. In Emerging frameworks and methods: Proceedings
of the Fourth International Conference on Conceptions of
Library and Information Science (CoLIS 4), pages 137–
150, Greenwood Village, CO. Libraries Unlimited.
Bernstein, M. S., Kleek, M. V., mc schraefel, and Karger,
D. R. (2007). Management of personal information
scraps. In Rosson, M. B. and Gilmore, D. J., editors,
CHI Extended Abstracts, pages 2285–2290. ACM.
Bettoni, M. C., Ottiger, R., Todesco, R., and Zwimpfer, K.
(1998). Knowport: A personal knowledge portfolio tool.
In Reimer, U., editor, PAKM, volume 13 of CEUR Work-
shop Proceedings. CEUR-WS.org.
Blackler, F. (1995). Knowledge, knowledge work and orga-
nizations: An overview and interpretation. Organization
Studies, 16(6):1021–1046.
Boettger, M. (2005). Pkm and “cues to knowledge”. Tech-
nical report, 25Uhr.de.
Bontas, E. P., Tempich, C., and Sure, Y. (2006). Ontocom:
A cost estimation model for ontology engineering. In
Cruz, I. et al., editors, Proceedings of the 5th Interna-
tional Semantic Web Conference (ISWC 2006), volume
4273 of Lecture Notes in Computer Science (LNCS),
pages 625–639. Springer-Verlag Berlin Heidelberg.
Cleverdon, C. W., Mills, L., and Keen, M. (1966). Factors
determining the performance of indexing systems. Tech-
nical report, ASLIB Cranfield Research Project, Cran-
field.
Cutrell, E., Dumais, S. T., and Teevan, J. (2006). Searching
to eliminate personal information management. Com-
mun. ACM, 49(1):58–64.
Davenport, T. H. (2005). Thinking for a Living: How to
Get Better Performances And Results from Knowledge
Workers. Harvard Business School Press.
ICEIS 2008 - International Conference on Enterprise Information Systems
104
Despres, C. and Chauvel, D. (2000). Knowledge Hori-
zons: the present and promise of Knowledge Manage-
ment. Butterworth-Heinemann.
Drucker, P. F. (1985). Management: Tasks, responsibilities,
practices (Harper & Row management library). Harper
& Row.
Esser, K. B. (1998). Ein Modell zur Verknüpfung des per-
sönlichen Gedächtnisses mit externen Informationsspe-
ichern. PhD thesis, Freie Universität Berlin.
Feldman, S., Duhl, J., Marobella, J. R., and Crawford, A.
(2005). The hidden costs of information work. Technical
report, IDC.
Frand, J. and Hixon, C. (1999). Personal knowledge man-
agement : Who, what, why, when, where, how? Speech.
working paper.
Glushko, R. (2006). 3. information organization and,or,vs
search. Lecture Note.
Groza, T., Handschuh, S., Möller, K., and Decker, S.
(2007). Salt - semantically annotated latex for scientific
publications. In Franconi, E., Kifer, M., and May, W.,
editors, ESWC, volume 4519 of Lecture Notes in Com-
puter Science, pages 518–532. Springer.
Gruber, T. R. (1989). The acquisition of strategic knowl-
edge. Academic Press Professional, Inc., San Diego, CA,
USA.
Higgison, S. (2005). Your say: Personal knowledge man-
agement. Insight Knowledge, 7(7).
Iske, P. and Boekhoff, T. (2002). The value of knowledge
doesn’t exist. In Karagiannis, D. and Reimer, U., edi-
tors, PAKM, volume 2569 of Lecture Notes in Computer
Science, pages 632–638. Springer.
Jones, W. and Bruce, H. (2005). A report on the nsf-
sponsored workshop on personal information manage-
ment. report.
Jones, W., Bruce, H., and Dumais, S. (2001). Keeping
found things found on the web. In CIKM ’01: Proceed-
ings of the tenth international conference on Information
and knowledge management, pages 119–126, New York,
NY, USA. ACM Press.
Klyne, G. and Carroll, J. J. (2004). Resource de-
scription framework (RDF): Concepts and ab-
stract syntax. http://www.w3.org/TR/2004/
REC-rdf-concepts-20040210/.
Krötzsch, M., Vrandecic, D., and Völkel, M. (2006). Se-
mantic mediawiki. In Cruz, I., Decker, S., Allemang,
D., Preist, C., Schwabe, D., Mika, P., Uschold, M., and
Aroyo, L., editors, Proceedings of the 5th International
Semantic Web Conference (ISWC06), volume 4273 of
Lecture Notes in Computer Science, pages 935–942,
Athens, GA, USA. Springer.
Lethbridge, T. (1998). Metrics for concept-oriented knowl-
edge bases. International Journal of Software Engineer-
ing and Knowledge Engineering, 8(2):161–188.
Lethbridge, T. C. (1991). A model for informality in
knowledge representation and acquisition (an extended
abstract). presented at Workshop on Informal Comput-
ing, May 29-31 1991, Santa Cruz CA.
Likert, R. (1932). A technique for the measurement of atti-
tudes. s.n., New York.
Ludwig, L. (2005). Semantic personal knowledge manage-
ment. Technical Report D11.01_v0.01, DERI Galway.
Maier, R. (2004). Knowledge Management Systems: In-
formation and Communication Technologies for Knowl-
edge Management. Springer.
Maier, R. and Schmidt, A. (2007). Characterizing knowl-
edge maturing: A conceptual process model for integrat-
ing e-learning and knowledge management. In Gronau,
N., editor, 4th Conference Professional Knowledge Man-
agement - Experiences and Visions (WM ’07), Potsdam,
volume 1, pages 325–334, Berlin. GITO.
Maurer, H. (1999). The heart of the problem: Knowl-
edge management and knowledge transfer. In Proc. EN-
ABLE’99, pages 8–17. Espoo-Vantaa Institute of Tech-
nology.
Mitchell, A. (2005). The rise of personal km. Inside Knowl-
edge, 9(1).
Nonaka, I. and Takeuchi, H. (1995). The Knowledge-
Creating Company : How Japanese Companies Create
the Dynamics of Innovation. Oxford University Press.
North, K. (2007). Produktive wissensarbeit. In 5. Karl-
sruher Symposium für Wissensmanagement in Theorie
und Prxais. CD-ROM.
Oren, E., Völkel, M., Breslin, J. G., and Decker, S. (2006).
Semantic wikis for personal knowledge management.
In Database and Expert Systems Applications, volume
4080/2006, pages 509–518. Springer Berlin / Heidel-
berg.
Pédauque, R. T. (2003). Document: Form, sign and
medium, as reformulated for electronic documents.
Pirolli, P. and Card, S. K. (1995). Information foraging in
information access environments. In CHI, pages 51–58.
Polanyi, M. (1958). Personal Knowledge: Towards a Post-
Critical Philosophy. Routledge & Kegan Paul Ltd, Lon-
don.
Teevan, J., Alvarado, C., Ackerman, M. S., and Karger,
D. R. (2004). The perfect search engine is not enough:
a study of orienteering behavior in directed search. In
CHI ’04: Proc. of the SIGCHI conf. on Human factors in
computing systems, pages 415–422. ACM Press.
Toms, E. G. (2000). Serendipitous information retrieval. In
DELOS Workshop: Information Seeking, Searching and
Querying in Digital Libraries.
Van Rijsbergen, C. J. (1979). Information Retrieval, 2nd
edition. Dept. of Computer Science, University of Glas-
gow.
Varian, H. R. (1999). The economics of search. In SIGIR
’99: Proceedings of the 22nd annual international ACM
SIGIR conference on Research and development in in-
formation retrieval, page 1, New York, NY, USA. ACM.
Völkel, M. (2007). A semantic web content model and
repository. In Proceedings of the 3rd International Con-
ference on Semantic Technologies.
Vrandecic, D. (2006). Ontology evaluation for the web -
phd proposal. In Diederich, J., Motta, E., and Bontas,
E. P., editors, Proceedings of the KnowledgeWeb PhD
Symposium KWEPSY 2006, Budva, Montenegro.
COST-BENEFIT ANALYSIS FOR THE DESIGN OF PERSONAL KNOWLEDGE MANAGEMENT SYSTEMS
105