A Framework for Ontology-Based Knowledge Management System

1

• A Framework for Ontology-Based Knowledge
  Management System
• Jiangning Wu
• Dalian University of Technology, China

2
Introduction

• Introduction
• Background
• Problems
• Solution
• Focus
• Contributions

3
Background

• The goal of a general KMS is to provide the right
  knowledge to the right people at the right time
  and in the right format.
• Through KMSs, users can access and utilize the
  rich sources of data, information and knowledge
  stored in different forms.

4
Problems

• Traditional KMSs are based on the existing data
  repositories and users needs.
• For knowledge discovering, users submit queries
  to the system and receive knowledge by keyword
  match.
• But keyword-based systems cannot understand the
  meaning of data. They are inflexible and stifle
  for knowledge creation.

5
Solution

• The emerging ontology-based KMSs can find the
  content-oriented knowledge that people really
  want.
• The domain ontology is powerful in knowledge
  representation and associated inference.

6
Focus

• We mainly focus on performing the activity for
  projects and domain experts matching.
• In project management, it is not easy to choose
  an appropriate domain expert for a certain
  project if experts research areas and the
  contents of the projects are not understood very
  well.

7
Contributions

• Our contributions are describing experts
  research areas and the contents of the projects
  by separated ontologies based on the same
  standard subject category of China.
• So the matching problem is transformed into
  calculating the semantic similarities between
  ontologies.

8
Contributions

• To calculate the similarity between documents, we
  propose an integrated method based on node-based
  method and edge-based method to solve this
  problem.

9
Ontology in KR

• Ontology in Knowledge Representation
• Ontology in General
• T.R. Gruber
• Why Ontology
• Our Ontology

10
Ontology

• Research on knowledge representation has been a
  focus of AI and IS disciplines for a number of
  years.
• Much of contemporary research extends the seminal
  work within AI discipline, of which research in
  ontology has been one of the beneficiaries.

11
Ontology

• Research in computational ontology has
  traditionally sought to develop structure for the
  purpose of knowledge subsumption.
• The goal of such research aims to develop
  generic, reusable representations of domain
  ontology.

12
T.R Gruber

• T.R. Gruber claimed An ontology is an explicit
  specification of a conceptualization. The term is
  borrowed from philosophy, where an ontology is a
  systematic account of existence.
• For knowledge-based systems, what exists is
  exactly that which can be represented.

13
Ontology

• An ontology in short is an explicit description
  of a domain
• concepts
• properties and attributes of concepts
• constraints on properties and attributes
• Individuals (often, but not always)
• An ontology defines
• a common vocabulary
• a shared understanding

14
Why Ontology

• To share common understanding of the structure of
  information
• among people
• among software agents
• To enable reuse of domain knowledge
• to avoid re-inventing the wheel
• to introduce standards to allow interoperability

15
Why Ontology

• To make domain assumptions explicit
• easier to change domain assumptions (consider a
  genetics knowledge base)
• easier to understand and update legacy data
• To separate domain knowledge from the operational
  knowledge
• re-use domain and operational knowledge
  separately (e.g., configuration based on
  constraints)

16
Our Ontology

• The ontology is a collection of concepts and
  their relationships, and serves as a
  conceptualized vocabulary to describe an
  application domain.
• In our study, it is created by means of Protege,
  which is developed by Stanford University.

17
Our Ontology

• The initial concepts in our ontology are broadly
  extracted from the standard subject category of
  China.
• To make the selected concepts more suitable for
  our concerned projects and domain experts, a tool
  called Concept Filler is developed, which is
  simply an interface to help domain experts assign
  proper concepts and weights manually.

18
Interface
19
Our Ontology

• When specifying the concept, the corresponding
  weight value ranging from 0 to 1 is also assigned
  to itself aiming to distinguish its importance.
• The relationships in an ontology are explicitly
  named which can reflect the context of the domain
  knowledge.

20
Relationships

• Many types of relationships can be found in
  ontology construction as we have known, such as
  IS-A relation, Kind-of relation, Part-of
  relation, Substance-of relation, and so on.
• Since IS-A (hyponym / hypernym) relation is the
  most common concern in ontology presentation,
  only this kind of relation is therefore
  introduced in our research for simplification.

21
Our Ontology
22
Matching Method

• Matching Method
• Node-based Method
• Edge-based Method
• Shortcomings
• Integrated Method

23
Considerations

• Calculating the similarity between concepts based
  on the complex relationships is a challenging
  work.
• Unfortunately no method can deal with the above
  problem effectively up to now.
• Considering some similarity calculation methods
  have been developed based on the simplest
  relation - IS-A relation, only this kind of
  relation is retained in our study.

24
Node-based Method

• Resnik used information content to measure the
  similarity.
• His point is that the more information content
  two concepts share, the more similarity two
  concepts have.

25
Node-based Method

• The similarity of two concepts c1 and c2 is

Considering many inherited concepts may have more
than one senses, the above formula is modified as
26
Edge-based Method

• Leacock and Chodorow summed up the shortest path
  length and converted this statistical distance to
  the similarity measure.

27
Shortcomings

• Both node-based and edge-based methods only
  simply consider two concepts in the same concept
  tree without expanding to two lists of concepts
  in different concept trees.
• However the fact is when we describe different
  documents in the same domain using ontology
  structures, homogeneous but heteromorphic concept
  trees are often formed.

28
Shortcomings

• The matching problem to be solved here is
  calculating the similarity between two different
  concept trees, not between two concepts in the
  same tree.
• So we have to develop a new method that can
  calculate the similarities between two lists of
  concepts in different trees, by which the
  quantified similarity value can show how similar
  the documents are.

29
Shortcomings

• The node-based method does not concern the
  distance between concepts.
• From the four-hierarchy concept tree, we can see
  that if concepts C21, C31 and C36 have the same
  sense and the equal frequency, we may get the
  following result according to the node-based
  method
• sim(C21, C31) sim(C21, C36)

30
Shortcomings

• However, it is obvious to see that concepts C21
  and C31 are more similar since C31 is the direct
  inheritor of C21.

31
Shortcomings
32
Shortcomings

• In contrast to the node-based method, the
  edge-based method only considers the
  relationships between concepts and ignores the
  weights of concepts.
• Both concepts C31 and C32 respectively have only
  one edge with C21. According to the edge-base
  method, the same similarity value can be
  obtained.

33
Shortcomings

• But, if C31 has bigger weight than C32, C31 is
  considered to be more important and the
  corresponding similarity value between C31 and
  C21 should be greater.

34
Integrated Method

• Before conducting the proposed method, the
  documents related to projects and domain experts
  should be formalized first that results in two
  vectors containing the concepts with their
  frequencies.

35
Integrated Method

• The similarity between cis and cjt

• The modified similarity

36
Integrated Method

• The similarity between two documents

37
Framework

• Ontologies Building
• Documents Formalization
• Similarity Calculation
• User Interface.

38
Framework
39
Evaluation

• Two measures to verify our ontology-based KMS

40
Evaluation

• Precision

41
Evaluation

• Recall

42
Conclusions

• An ontology-based method to match projects and
  domain experts is presented.
• The prototype system we developed contains four
  modules Ontology building, Document
  formalization, Similarity calculation and User
  interface.

43
Conclusions

• We discuss node-based and edge-based approaches
  to computing the semantic similarity, and propose
  an integrated approach to calculating the
  semantic similarity between two documents.
• The experimental results show that our
  ontology-based KMS performing the activity for
  projects and domain experts matching can reach
  better recall and precision.

44
Future Works

• As mentioned previously, only the simplest
  relation IS-A relation is considered in our
  study.
• When dealing with the more complex ontology whose
  concepts are restricted by logic or axiom, our
  method is not powerful enough to describe the
  real semantic meaning by merely considering the
  hierarchical structure.

45
Future Works

• So the future work will be focused on the other
  kinds of relations that are used in ontology
  construction.
• In other words, it will be an exciting and
  challenging work for us to compute the semantic
  similarity upon various relations in the future.

46

• THANKS