Ontology and CBR-based Dynamic Enterprise

Knowledge Repository Construction

Huiying Gao
School of Management and Economics, Beijing Institute of Technology, Beijing, China, 100081
Email: [email protected]

Xiuxiu Chen
School of Management and Economics, Beijing Institute of Technology, Beijing, China, 100081
Email:[email protected]



Abstract- The efficiency of knowledge sharing and learning
is the key to obtain sustainable development for the
knowledge-intensive industry. However, current application
of enterprise knowledge repository can hardly adapt to the
personalized retrieval with semantic expansion and can not
support the dynamic mechanism of knowledge sharing. This
paper focuses on an integrated framework and operating
processes of dynamic knowledge repository construction.
Through analyzing the key technology points of business
logic processing layer and data services layer particularly,
the ontology and CBR-based knowledge storage and
retrieval mechanism are studied, which improve the
effectiveness of knowledge management.

I ndex Termsontology; case based reasoning; knowledge
repository; semantic retrieval

I. INTRODUCTION
With the development of knowledge economy and the
superheating competition of the market, many
knowledge-intensive industries such as aviation,
advanced manufacturing, IT and consulting are suffering
from the distress of increasing knowledge assets outflow,
where technical requirements span is wide and
management task is difficult and complicated. The
recent research paper in the Journal of Knowledge
Management shows that knowledge application is
directly related to organizational performance [1]. A good
environment of knowledge sharing and learning will
improve the enterprises efficiency and service; finally
make it get sustainable competition.
However, the most precious knowledge of the
enterprise often exists in the minds of its employees,
work processes, experiences, and in electronic or written
form, etc. The research paper published in the
Information Systems Research in 2010 pointed that
knowledge capabilities with IT contribute to firm
innovation [2], and also demonstrated that individuals
will absorb and recreate knowledge in the organization
with strong knowledge storage and retrieval capabilities
[3].Therefore, the knowledge repository construction will
play a vital role in the knowledge transformation and
sharing between personal and the enterprise.
The massive domain knowledge resources and
experiences are accumulated in many
knowledge-intensive enterprises in recent years. However,
there are still many urgent problems that they have to
face. There is lack of general knowledge model for
domain support, and there is lack of semantic support for
the knowledge case representation, retrieval and reuse.
The organization of knowledge repository fixes in a
single form and the hierarchical structure is ambiguous.
Besides, the knowledge repository with weak case
learning ability cannot meet the dynamic mechanism of
knowledge reuse.
This paper aims at a dynamic enterprise knowledge
repository construction on ontology and case-based
reasoning (CBR). At first in section II the ontology and
CBR are briefly introduced and the state of art is
summarized. After that in section III the framework of
ontology and CBR-based dynamic enterprise knowledge
repository is proposed and the key technology points of
business logic processing layer and data services layer are
described respectively. Take an information system
consulting company as an example, in section IV, the
process of the enterprise domain ontology is shown and
the dynamic mechanism of case repository is designed in
section V, including the case representation, organization,
retrieval and learning. Finally section VI displays our
conclusion and outlook.
II. ONROLOGY AND CASE-BASED REASONING
A. Ontology
Ontology is the formal clear specification of the
sharing conceptual model; it captures the basic domain
terms and their relationships, defines the relevant rules to
determine the vocabulary extension, and finally forms a
knowledge structure model in specific areas in order to
achieve the consistent understanding of the domain
knowledge [4]. As ontology provides a clear semantic
This work was supported in part by the National Natural Science
Foundation of China under Grant 71102111 and Beijing Institute of
Technology under Grant 3210050320908.
Corresponding authorHuiying Gao Email: [email protected]
JOURNAL OF SOFTWARE, VOL. 7, NO. 6, JUNE 2012 1211
2012 ACADEMY PUBLISHER
doi:10.4304/jsw.7.6.1211-1218


and knowledge description of the concepts and
interrelation, it can be adapted to the cases despription
and hierarchical structure storage and support the
semantic knowledge retrieval.
B. Case-based Reasoning
CBR is an important problem solving and learning
method based on knowledge in the artificial intelligence
field. It has good extensibility and the ability to learn [5].
Each processed problem is described as the feature set
and solutions, then stored as a case in the system.When
the new problem comes, the most similar case will be
retrieved and modified if necessary. The modified cases
will be seen as an new case and stored in order to realize
the reuse and relearning of the cases. Also , case retrieval
will be the key point of the case reasoning.
C. Related Research
Dynamic knowledge repository management is a hot
issue in the current information field. Many scholars have
done a great deal of researches based on ontology.
In the Journal of Information Science, many related
researches are delivered such as knowledge extraction [6],
the uniform knowledge representation [7], the knowledge
matching and retrieval [8] with ontology technology and
so on. Moreover, in the Journal of Knowledge
Management, information processing based on ontology
in the construction process of the knowledge management
system is explored [9]. Besides, Liao Liangcai imported
ontology into the knowledge management system and
realized the enterprise knowledge management through
the semantic expansion, reasoning and retrieval finally
[10]. All these researches show the important role of
ontology in the realization of knowledge sharing and
reuse.
Meanwhile, the unstructured knowledge such as
experiences and minds of employees is more suitable to
store in the form of case, and it is easier to realize the
dynamic knowledge management based on CBR. So the
following research focuses on dynamic enterprise
knowledge repository model with the effective
combination of ontology and CBR technology. With the
establishment of domain ontology, the semantic
consistency for knowledge representation and storage is
ensured and semantic expansion of the users query
demand is realized. Besides, case modification, learning,
reuse and new case formation enhance the adaptability of
knowledge repository and realize its dynamic
construction.
III. ONTOLOGY AND CBR-BASED DYNAMIC ENTERPRISE
KNOWLEDGE REPOSITORY FRAMEWORK
A. Integrated Framework
An ontology and CBR-based dynamic enterprise
knowledge repository model is discussed in this part. The
basic framework is illustrated in Fig.1, which adopts a
four-tier system structure: customer application layer,
business logic processing layer, data services layer and
system layer. All these different layers work closely
together to complete the work of knowledge management.
The main functions of each layer are described as
follows.

WWW
System
Administrator
Normal
Users
Knowledge
Engineer
Business
Logic
Processing
Layer
System
Management
Module
Case
Retrieval
Processing
Module
Case
Learning
Module
Multimedia
Library
DBMS
OS
OS
Case
Reasoning
Matching
Module

Knowledge
Acquisition
Module
User
Information
Database
Case
Modify
Module
Ontology Management Module
Customer
Application
Layer
Data
Services
Layer
System
Layer
Case
Definition
Module
Ontology
Database
Case
Database
Server
Communication
Protocols

Figure1. The framework of dynamic knowledge repository
The customer application layer provides a good
interaction interface for the users including knowledge
engineers, normal users and the system administrator.
The business logic processing layer encapsulates the core
functional modules of the knowledge repository system,
which are responsible for knowledge acquisition,
representation, case definition and storage, ontology
analysis, as well as case retrieval, learning and so on. The
data service layer is the basic part of this system, which
logically realizes the expression and organization of the
user information database, multimedia database, ontology
database and case database. The system layer aims at
offering operation system, database management system,
Server, data standards, network, communication
protocols, and many other physical supports.
As the three types of users have different functional
requirements, the operational processes will been
analyzed and illustrated respectively as follows.
Generally speaking, knowledge engineers firstly
acquire explicit and tacit knowledge from related experts,
enterprises original nonstructural database and many
other channels with the knowledge acquisition module.
Secondly, the core domain concepts are extracted and
enterprise ontology database is constructed and
maintained by the ontology management module.
According to the hierarchical structure of ontology,
knowledge engineers annotate the semantic information
and definition for the cases, and build case classification
index. Finally, the metadata of the defined cases is stored
in the case base while non-structured data or the original
documents are stored in multimedia library with the XML
format.
Normal users follow a different process. Firstly,
several key words are input through the case retrieval
processing module, and then semantic annotation will be
added based on the ontology, the users profile and the
retrieval history. Then the users query can be represented
by the semantic vector, which will be matched with
source cases later. Finally, cases beyond a certain
threshold are sent to the client application layer.
Sometimes, failure cannot be avoided. When the users
needs fail to be met, some cases will be combined
together and refined according to ontology by case
modify module. In this way, a new case will be formed
and stored in case database through case learning module,
1212 JOURNAL OF SOFTWARE, VOL. 7, NO. 6, JUNE 2012
2012 ACADEMY PUBLISHER


while other additional information will be added to the
multimedia library. It can be seen obviously that ontology
management module provides semantic support through
the whole process of dynamic management.
System administrator realizes the management of
different user permission with reference to the users
records in the user information database through the
system management module.
As business logic processing layer and data service
layer play an important role in the dynamic knowledge
repository system, they will be discussed further in the
part B and part C.
B. Business Logic Processing Layer
The business logic processing layer of the dynamic
knowledge repository system put forward in this paper
includes the following main function modules.
1) System Management ModuleIt is responsible for
controlling the access authorities of the different types of
users. For example, the administrator manages users
access; knowledge engineers deal with knowledge
management and maintenance, while normal users
retrieve knowledge.
2) Knowledge Acquisition ModuleTwo ways can be
adopted to acquire domain knowledge. One approach is
to arouse the domain experts initiatives to obtain
knowledge through the brain storm or Delphi method,
which is simple, direct and efficient. But certain
dependence on the experts will be the main problem. The
other way is to dig for knowledge from the present
existing results, including the enterprise non-structural
database, internal materials, the patent documents,
intranet, extranet, BBS and other internal communication,
which will be the preparation for the ontology
construction.
3) Ontology Management Module Based on the
analysis of the domain scope and characteristics,
knowledge engineers extract the core concepts as well as
their respective sub-categories from the enterprise. Also,
the attributes of each sub-category and the constraints of
the attribute are also given. Finally, architecture of
ontology will be set up with various interrelationships,
such as project classification system, enterprise business
department classification system and so on. All these
domain ontology will form ontology modeling and be
stored in ontology database. The detail will be illustrated
in the section IV.
4) Case Definition ModuleWith reference to the
ontology model, some key features are extracted from the
documents, the existing project cases and long-term
accumulated experiences, which are used to express the
corresponding cases in the particular way. Additionally,
the ontology- based case classification index mechanism
will be established in order to organize the cases clearly.
5) Case Retrieval Processing ModuleThe function of
this module is to segment and analyze the users query,
and then extend to the different extracted words. The
words in user queries can directly map to the concepts,
attributes or the instances of ontology, with which every
dimension of query vector can be replaced in the
semantic view.
6) Case Reasoning Matching ModuleThe function of
this module is to achieve the semantic matching between
the users query and the annotated cases. Much work is to
expand synonym concept and clarify the ambiguous
query information or intension. Based on the results of
users query pre-processing, query vector with semantic
expansion is generated for the semantic matching with
the representation vectors of source cases in the case
database. After computing the similarity between them
with certain retrieval strategy and algorithm, cases
beyond certain threshold will be returned in order.
7) Case Modify ModuleAs the scale of the knowledge
repository is so limited that it cannot satisfy all users
demands, the module provide the function of case
combination and adjustments according with ontology
defined to form new cases.
8) Case Learning ModuleThe function of this module
is to learn the modified cases automatically according to
certain rules and enrich the case database gradually.
Specific information will be described in section V .
C. Data Services Layer
According to the category of enterprise knowledge,
they are stored in the User Information Database,
Ontology Database, Case Database and Multimedia
Library.
User information database is designed to store all the
users' personal background information, retrieval history
and reuse records. The prescriptive documents and
relational database reflect users needs and preferences in
order to improve the pragmatic retrieval.
The ontology database is suggested to maintain the
domain concepts, the properties, the attribute constraints
and relations between these concepts, and finally form the
concept model with clear structure. Ontology is so
important that many other parts of the knowledge
repository system are established based on it. It
contributes to the domain knowledge reasoning, case
retrieval, matching and learning for dynamic knowledge
repository management.
The case database is prepared to store the accumulated
cases of the enterprise for a long term. It aims at offering
reference for the subsequent ones, which is vital to realize
knowledge reuse effectively.
Multimedia library stores semi-structural even
non-structured data such as the corresponding document
of project cases, related project design, the flow chart,
technology and method, the source code, photos, and
video conference and so on.
IV. CONSTRUCTION OF ENTERPRISE ONTOLOGY
It is obviously that enterprise ontology is the basis of
the whole knowledge system and determines the
performance as well as the quality of the operation.
Therefore, how to establish enterprise ontology correctly,
effectively and logically is very important.
Based on the dynamic knowledge repository
framework above, we adopt the framework method [4]
JOURNAL OF SOFTWARE, VOL. 7, NO. 6, JUNE 2012 1213
2012 ACADEMY PUBLISHER


given by Uschold & Gruninger to build the enterprise
ontology with ontology modeling tools protg 3.3.4.
Taking an information system consulting company as an
example, Fig. 2 and Fig. 3 illustrate the hierarchy and the
fragment of the ontology respectively. Firstly, according
to the outline, we identify enterprise domain ontology
scope in many ways with knowledge engineers and other
experts, such as research, interview, brainstorming and so
on. Secondly, the core domain concepts are extracted
after the analysis and evaluation, and the interrelation and
hierarchical structure are defined as well. The main
relationships of the ontology may be <is A>, <a part of>,
<equal to>, <similar to>, <instance of> and <attribute of>.
Fig.2 shows synonymous relationship <equal to>and
hyponymy such as <subclass of>. Finally, the attributes
of each sub-category and the constraints of the attribute
are also given and the ontology is described with OWL.
A small section of the source code with OWL is shown
in the following, which is generated automatically by the
ontology modeling tools protg 3.3.4 after the ontology
is set up. As we can see, Video conference system is
equal to Video session system, and Multi-media
project is the subclass of Project.
</owl:Class>
<owl:Class rdf:ID="Video_conference_system">
<owl:equivalentClass
rdf:resource="#Video_session_system"/>
<rdfs:subClassOf
rdf:resource="#Multi-media_project"/>
</owl:Class>
</owl:Class>
<owl:Class rdf:ID=" Multi-media_project ">
<rdfs:subClassOf rdf:resource="#Project"/>
</owl:Class>
Furthmore ,object properties are defined as follows
<owl:ObjectProperty rdf:ID="has_name"/>
<owl:ObjectProperty rdf:ID="has_budget"/>
<owl:ObjectProperty rdf:ID="has_constructors"/>
<owl:ObjectProperty rdf:ID="has_goals"/>
<owl:ObjectProperty rdf:ID="has_owner"/>
<owl:ObjectProperty rdf:ID="has_requirement"/>
As ontology exerts strong influences on the utilization
of knowledge repository, the evaluation method based on
the feedbacks of users is adopted in the domain ontology
assessment. That is to say, knowledge engineers analyze
the richness of ontology information (including concepts,
attributes, and the definition of an example) and semantic
intensity (including ontology structure and relationships)
based on the basis of users satisfaction.
Except the evaluation, domain ontology maintenance is
also necessary. It mainly refers to a series of adjustments,
error corrections, perfect and adaptability maintenance
work. Correction maintenance focuses on put right in the
use. Perfect maintenance refers to ontology expansion
work with the increase of knowledge. Adaptive
maintenance points to the refreshment of the structure,
attributes and relationships with the changes of
environment. With tracking and management of ontology
model continually, it will provide better support for the
dynamic knowledge repository.

Figure 2. Ontology hierarchy of project types


Figure 3.Ontology fragment of project types
VONTOLOGY-BASED CASE DATABASE CONSTRUCTION
A. Case Representation
Actually, case representation is a kind of knowledge
expression, which code knowledge into a set of data
structure for computer. In this paper the case database is
described as: {case , case , ..., case , ,..., }; 1 2

and
case

is illustrated with ordered pairs <case ID, case title,
initial problem description, solutions description,
additional information>.Let we take

=ID,TI,IN,SO,AD>
for
short. The unique identifier of one case is expressed
with ID . SO refers to the case solution and the
specification, including performance, causes, the main
problems, the economic and social benefits, which will be
an significant basis of case reuse. AD is related with
corresponding documents and multimedia content.
Document element and other unstructured data are more
suitable to store in the form of multi-media format, which
is convenient for the user to understand the whole case.
We divide the case retrieval into case title part
represented byTI and initial problem description part
represented by IN , which are the basis of the case
retrieval. With the consideration of ontology database,
1214 JOURNAL OF SOFTWARE, VOL. 7, NO. 6, JUNE 2012
2012 ACADEMY PUBLISHER


feature vector is extracted from these two parts through
analysis and expressed by
1 2
i i i , im m
C { ,w ( t ,w )},i n.
. .
= =
1 1 2 2
t ,w t

and
ij
t represents
th
j feature of
th
i
case,
ij
W is the
corresponding weight of the
th
j
feature j m s s 1 .
B. Case Organization
In order to improve the speed of case retrieval, the
system organizes the cases in the hierarchical structure
according to case classification ontology. Then index
mechanism is built on the different fields, such as the
project classification or industry index, as shown in Fig.4.
So, when the user input with some search features,
appropriate index is adopted to search the optimized
cases efficiently with sematic expansion.

Project
E-business
project
E-government
project
Multi-media
project
Network
Construction
project
Enterprise
management
...
Gold
Project
Government
Portals
Golden
Finance
Project
Golden
Tax
Project
Video
Conference
system
Video
Surveillance
system
Distance
Education
OA
ERP
...
...

Figure4. The case organization based on project types
C. Case Retrieval
The case retrieval is conducted with the semantic
expansion based on the enterprise ontology. The reason
the semantic query expansion with ontology is that in
query language there are several situations as follows.
Firstly, there are many synonyms. It is quite common
in natural language, for example E-commerce
and E-business both mean commerce conducted
electronically. The relationship between these words is
called synonyms in the ontology. Besides, sometimes
users prefer taking some well-known words for short, for
example, online-offline is often replaced with OO for
short. So, when one user enters some key words, it can be
extended by ontology to its synonym.
Secondlythere are many concepts with ambiguities
and pragmatic environmental differences. In many cases,
the phenomenon of polysemy appears. Take the
word project as an example, it may mean engineering
in the broad perspective, or scientific research in the
scientific field, while sometimes it means program or
planning. Besides, when user input finance as his
application industry demand, it may be considered as the
government industry or financial industry. If a record of
the user's personal information is e-government
consultant and many of historical records are related with
the government program rather than financial industry
project, we will prefer to define the application industry
as government. Thus, in order to eliminate the ambiguity,
we should consider users background information firstly
to clarify its specific and tacit implications
Thirdly, there are some words of superordinate and
subordinate concept. In many cases, only through this
relationship can we retrieve the potential information,
such as project, multimedia and video conference.
According to the case representation method illustrated
in the part A of this section, suppose there are
m keywords, and one semantic vector can be expressed
as
1 2
, 1, 2,...,
i i i im
T t t t i n
T
( = =

.
Based on semantic expansion with the domain
ontology, each of the key word can be expressed as a
semantic vector, which will be used to match with the
keywords defined in the source cases. According to the
description in the part A of this section, as the importance
between case title part and initial problem description
part will be different in concrete applications, we should
consider the weight of one keyword appeared in the title
part and initial problem description part. Let
o
represent
the weight of the title part; let
|
represent the weight of
initial problem description part and
1 o | + =
. Generally
speaking, the main content will be shown more obviously
in the title of one case, so we define
o | >
in the case
retrieval process of this paper.
Suppose C as a candidate case set, and i
C
eC is one
candidate case of the candidate case set,
then
*
C
represents the query vector that is consistent with
the users demand. We define the feature vector of a case
expressed as
1 2
,
i i i im
T t t t
T
( =

. So the semantic
similarity between
C
i
and
*
C
can be defined as (1).
* * *
( ) ( ) ( )
( , , + = )
i i i
SIM C C SIM C C SIM C C
TI IN
and
o |
o | o |
= +
> 1

In the formula above
*
( )
i
SIM C C
TI
means the
similarity of the title part and
*
( )
i
SIM C C
IN
means
similarity of the initial problem description part.
While computing the value of
*
( )
i
SIM C C
TI
, the
frequency of each keyword in the vector space in every
case title is ignored. Therefore, if
th
j keyword doesnt
exist in the title of the case i
C
, in another word,
th
j
keyword isnt important in the title of the case i
C
, and
then
ij
t = 0. Else,
th
j keyword is very important in the
title of the case
C
i , then ij
t = 1
.
However, while computing the value of
*
( )
i
SIM C C
IN

,
the frequency of each keyword in the vector space in
initial problem description part of every case should be
considered, as the descriptions of cases are more
complicated and can reflect much information. So, the
frequency of
th
j keyword will be numbered. In this part,
let
ij
t i , , ,n ; j , ,...m = = 1 2 1 2
denote the frequency
of
th
j keyword in the description of
th
i case and
m
is the
total number of the terms. So in the same
m
dimension
vector space the description of
th
i case can be represented
by
1 2
,
i i i im
T t t t
T
( =

. Then we can realize that value
of ij
t
denotes the importance of
th
j keyword.
As
ij
t > 1, the following step is to regress the feature
vector of the case description to [0, 1] for convenient
calculation. Let
(1)
JOURNAL OF SOFTWARE, VOL. 7, NO. 6, JUNE 2012 1215
2012 ACADEMY PUBLISHER


{ }
{ } { }
1
1 1
min
max min
ij ij
j m
ij
ij ij
j m j m
t t
d
t t
s s
s s s s

In this way, not only the title part but also the case
description part can be represented in the following
expression 1 2
,
i i i im
D d d d
T
( =

. Particularly in the title part
as the value of ij
t e|0,1|
, then 1 2
,
i i i im
D d d d
T
( =

is
the same as
1 2
,
i i i im
T t t t
T
( =

.
Set the users query vector as standard, and then the
query vector space should be
*
( , ,... ,..., )
j m
C W I w w w w
T
= - =
1 2
I represents a unit matrix. W
T
is set with the scoring
method. Only the user can understand his demand most
clearly, therefore,
j
W
T
is calculated by the keywords
preferences of the user. We divide the users keyword
preference level into the following fuzzy set:{very
important, important, common, unimportant, very
unimportant}. For quantitative analysis, the fuzzy set can
be mapped into the vector{54321}. If the users
preference vector for the m keyword
is m j
( x , x ,..., x ) x , < <
1 2
1 5
and integer, then
=
j
j
m
j
j
w
x
x
=

1

Therefore, the similarity between the query
information and source case information in different part
is calculated with 4and5respectively.

( ) cos( , )
* *
m
j
j
i i
SIM C T
TI
C w C
=
=

*
*
( , ) COS( , )
d
* *
*
*
m
IN j
j
m
j
j
m
ij
m
j
j
m m
j
ij j
j j
i i
i
i
j
SIM
w
c
w
C C w D C
D C
D C
d c
=
=
=
=
= =
=

1
1
1
1
2 2
1 1

While computing the similarity of query vector
*
C
and
semantic vector
0
C
1
C
n
C
,remember to represent the
largest similarity as max
*
( )
i
C C SIM
.If the value is larger
than or equal to the given thresholdc , the case will be
added into the search results. Finally, the search results
sort according to the similarity and be sent to the user.
D. Case Reuse and Learning
Generally speaking, the scale of the initial case base is
so limited that cannot satisfy all the different needs of
customers, so the dynamic management mechanism
becomes urgent. It mainly includes the realization of the
case reuse, modification and case learning on the basis of
case retrieval. Fig. 5 shows the flow chart of this dynamic
mechanism.
In the process of case reuse, two main problems should
be considered. One is the differences between the new
problem and the retrieval results. The other is which part
of the result can be reused. We define those cases met the
users demand group in the matching set expressed byG ,
that is
*
{ | ( ) }
i i
G C SIM C C c = > .
If
G u =
, then decompose the query vector by level in
accordance with ontology and calculate the similarity for
searching the matching set which is not u .Mark this
level as starting point, and combine the cases from down
to up with the ontology rules which can coordinate the
constraint and standard between different cases in the
process of combination. Two methods of combination are
appropriate: exhaustion and genetic algorithm. When the
number of case combination choices is small, that means
all possible combinations can be listed and the feasible
solution can be found out. We consider the one who got
the largest similarity with the new case as the optimal
solution. Otherwise, genetic algorithm will be more
efficient. Then, the process of case adjustment begins.
G=?
Decompose
query vector
Satisfied?
Learn
the case
Adjust
slightly
SIM ?
Merge
Cases
Store
Input query
vector
Start
End
Calculate
similarity
Match
sub-vector
?
Modify?
Combine
cases
Evaluate
case
Adjust
case
Y
N
Y
N
Y
Y
Y
N
N
N

Figure5. Flow chart of dynamic management
If G u = the process of case combination skips and
directly gets into the case adjustment process. As case
adjustment will also need to consider some other
information besides the case similarity, it should be
adjusted and decided according to case description,
solutions and other affiliated information with the help of
persons, in the manual or semi-automatic way. We define
the case after adjustment as target case, which is more
close to new problem than the source case.
The case is evaluated in practice, and then a new case
which meets our requirements will be learned. In order to
avoid the redundant information in the knowledge
repository, similarity between the new case and the
source case will be also calculated to determine the new
case will be stored or be processed further. Set certain
threshold
c
, S represents source case while N is new
case, so if ( , ) SIM S N s c , store the new case; else, merge
them together.
In this process, the ontology provides the semantic
support for the case retrieval, matching, combination,
adjustment and case learning.
E. Validation and Analysis
In order to verify the retrieval effectiveness, according
to the research work in this paper, a simple prototype
system of consulting industry is developed based on SQL
database and C # language.
Its difficult to make the case database rich at the very
beginning. Therefore, we mainly collect relevant data
from the enterprise original relational database system,
Intranet, and project materials and then sort into 40
(5)
(4)
(2)
(3)
1216 JOURNAL OF SOFTWARE, VOL. 7, NO. 6, JUNE 2012
2012 ACADEMY PUBLISHER


information system project consulting cases with analysis.
These cases are deposited into the knowledge repository
in the description and organization way illustrated above,
which can be seen as the preparation for the operation.
Suppose that a consultant who is mainly responsible
for e-government projects needs to retrieve some
information of history projects which provide project
management services for video conferencing system in
the finance field. Besides, if Ministry of Finance is main
project stakeholders, that will be better. So, the key
retrieval vector is extracted as Industry, Project Types,
Services and Main Project Stakeholders. This consultant
enters the following vector value:
, , Project , [ ]
T
ij
Finance Video conferencing system Management Ministry of Finance T =
The case retrieval interface is shown in Fig. 6.

Figure 6 User query interface
Also, his preferences for these keywords are defined
successively as important, very important, very important,
and unimportant respectively, which is shown in Fig.7. In
this way, the score could be described as
( , , , ) 4 5 5 2
and
got
=[1/4,5/16,5/16,1/8]
T
W
T
with (3). According to the
domain ontology above and the users information, the
retrieval vector can be expressed more precisely with
semantic and pragmatic expansion, which is shown in
detail in the Table. I. Particularly, the dimension of
finance is expressed as government rather than financial
industry, because the information of the consultant shows
that he is responsible for the E-government services and
the major project stakeholder is Ministry of Finance.

Figure 7 Weights determination interface
In this paper, we set
. , . 0 6 0 4 o | = =
the threshold
value of similarity . 0 5 c = then the consultants retrieval
vector was:
*
( / , / , / , / )
T T
C W I = - = 1 4 5 16 5 16 1 8 . Computing
with the retrieval algorithm and the expansion vector
shown above, cases those meet the formula
*
( )
i
SIM C C c >
are delivered as Fig.8.
TABLE I.
RETRIEVAL FEATURE VECTOR AND SEMANTIC EXPAND INFORMATION
ij
t

j
x

j
w

Vector Value synonymy Pragmatics Hyponymy
Industry 4 1/4 Finance
Government(Y);
Financial Industry(N)
Government
Project Types 5 5/16 Video conferencing system Video session system Multi-media
Services 5 5/16 Project Management PM; Supervisor Management
Main Project
Stakeholders
2 1/8 Ministry of Finance
Treasury Department
Finance Bureau

Person;
Organization


Figure8. The matching cases and similarities
Take the optimal matching case 7 as example. The
process is described as follows.
| |
| |
IN
=
( )=1
0.7559
( / , / , / , / )
*
,
, , ,
, / ,
*
( , )
*
( , ) 0.6 1 0.4 0.7559 = 0.9024
T
TI
SIM
TI
T
C C
T
D
SIM C C
SIM C C
T
T
=
=
=
= +
7
7
7
7
1 4 5 16 5 16 1 8
1 4 10 1
0 1 3 1 0

The traditional way of information retrieval is based on
keywords that the computer cannot understand users
potential semantic and personalized query demand.
Therefore, we conduct the semantic and pragmatic
retrieval research based on the domain ontology and
JOURNAL OF SOFTWARE, VOL. 7, NO. 6, JUNE 2012 1217
2012 ACADEMY PUBLISHER


users information. After the analysis, semantic and
pragmatic retrieval can get the best precision and recall
rate of the matching result among traditional retrieval
method, sematic and pragmatic method, as is shown in
TABLE II.
TABLE II.
THE COMPARISON OF THE RETRIEVAL METHOD

Retrieval
Results
Retrieval
Matching
Cases
Actual
Case
Number
Precision Recall
Traditional
searching
3 2 8 0.6667 0.375
Semantic
retrieval
5 4 8 0.8 0.5
Semantic
and
pragmatic
retrieval
8 7 8 0.875 0.875
VI. CONCLUSION
The dynamic knowledge repository construction is a
hot issue in the field of knowledge engineering at present.
This paper focuses on the construction method of
knowledge repository based on ontology and CBR. An
integrated framework is put forward and the operation
processes of dynamic knowledge repository system are
analyzed. With the construction of ontology the query of
the user can be expanded with more semantic information
and the case database can be formed more precisely. A
retrieval algorithm is designed and the dynamic
mechanism of the knowledge management is illustrated.
Moreover, taking a consulting company as the case study
background, the retrieval process is verified, which
proves the high accuracy and completeness of the
retrieval algorithm.
As the initial case samples are small and centralized,
further work should be done to conduct quantitative
analysis based on a large size and wide distribution of
case samples.
ACKNOWLEDGMENT
This work was supported in part by the National
Natural Science Foundation of China under Grant
71102111 and Beijing Institute of Technology under
Grant 3210050320908.
REFERENCES
[1] Annette M. Mills, Trevor A. Smith. Knowledge
management and organizational performance: a
decomposed view [J]. Journal of Knowledge Management,
2011, Vol.15 Iss1, pp.156171.
[2] K. D. Joshi, Lei Chi, Avimanyu Datta, Shu Han. Changing
the Competitive Landscape: Continuous Innovation
Through IT-Enabled Knowledge Capabilities
[J].Information Systems Research, Vol.21, No.3,
September 2010, pp.472-495.
[3] Chou, Shih-Wei. Knowledge creation: absorptive capacity,
organizational mechanisms, and knowledge
storage/retrieval capabilities [J].Journal of Information
Science, December 2005, vol.31, pp. 453-465.
[4] Stanford University Knowledge System Laboratory. A
translation approach to portable ontology specifications [R].
Stanford: Stanford University Knowledge System
Laboratory, 1992.
[5] Norbert Gronau, Frank Laskowski .Using Case- Based
Reasoning to Improve Information Retrieval in Knowledge
Management Systems [J].Springer-Verlag GmbH, 2003;
2663, pp.94-102.
[6] Wimalasuriya, Daya C., Dejing Dou. Ontology-based
information extraction: An introduction and a survey of
current approaches [J].Journal of Information Science,
June 2010, vol.36, pp.306-323.
[7] Wenhuan Lu, Ikeda, Mitsuru.A uniform conceptual model
for knowledge management of international copyright
law[J].Journal of Information Science, February
2008,vol.34,pp.93-109.
[8] Akbari, Ismail, Fathian, Mohammad. A novel algorithm
for ontology matching [J].Journal of Information Science,
June 2010, vol.36:pp.324-334.
[9] Chimay J. Anumba, Raja R.A. Issa , Jiayi Pan, Ivan
Mutis.Ontology-based information and knowledge
management in construction[J].Journal of Knowledge
Management,2008, Vol.8,Iss3, pp.218239.
[10] Liao Liangcai, Qin Wei, Shu Yu. Ontology-based Dynamic
Knowledge Management System [J].Computer
Engineering, 2009, 3516, pp.256-261.
[11] Gao Huiying, Yan Zhijun. CBR based Multi-agent System
Model for Case Retrieval of Information Systems
[J].Computer Engineering and Design, 2008, 29(5),
pp.1226-1228.
[12] Gao Huiying, Zhao Jinghua. Ontology-based Enterprise
Content Retrieval Method [J].Journal of Computers, 2010,
5(2), pp.314-321.
[13] Huiying Gao, Qian Zhu. Semantic Web based Multi-agent
Model for the Web Service Retrieval. Proceedings of
International Symposium on Computer Network and
Multimedia Technology, 2009.12, pp.897-900.






Huiying Gao, Dr. Associate professor, was born in
Shandong Province, China, in 1976. She received her
doctor degree in management science and engineering
from Beijing Institute of Technology in 2003. Now she is
an associate professor in the school of Management and
Economics, Beijing Institute of Technology. She has ever
been in Technical University of Berlin, Germany to do
her Ph.D. research work from 2002 to 2003. From 2008
to 2009 she has visited Karlsruhe Institute of Technology,
Germany for half a year. Her current research interests
include theory and method of information systems,
content and knowledge management, semantic retrieval,
intelligent information system etc. Ph: +86 (10)
68918830.



Xiuxiu Chen was born in Shandong Province, China,
1988. She is a Ph.D. degree candidate in the school of
Management and Economics, Beijing Institute of
Technology. Her research work includes information
management and knowledge management. Ph: +86 (10)
68918830.
1218 JOURNAL OF SOFTWARE, VOL. 7, NO. 6, JUNE 2012
2012 ACADEMY PUBLISHER