Transcript Presentation

Complex Adaptive
Knowledge
Management System
Supervisors:
Kurt April
(Knowledge Management)
Sonia Berman (Databases)
Anet Potgieter (Artificial Intelligence)
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Structure of Talk
–
–
–
–
–
Overview of project
Knowledge Engineering
Data Mining
Adaptive Presentation and Visualisation
Questions
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Project Overview
• “Organisations are information rich but knowledge
poor” – A Moore
• Knowledge Management:
– Leveraging Knowledge to create and sustain competitive
advantage.
• Project Challenge:
– Process information to create knowledge that optimizes
knowledge transfer to support effective knowledge
management.
• Strategy:
– Use a unique combination of database management,
artificial intelligence and visual representation to draw
useful concepts from vast quantities of information.
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Graphical illustration of roles as they fit into the KDD process
Presentation of
Knowledge to
User
Domain of
Knowledge
Engineering
Domain of
Data Mining
Domain of
Adaptive
Presentation &
Visualisation
Diagram adapted from The two Crows Corporation: Introduction to Data Mining and Knowledge Discovery. Third Edition
Graphical illustration of roles as they fit into the KDD process
Presentation of
Knowledge to
User
Domain of
Knowledge
Engineering
Domain of
Data Mining
Domain of
Adaptive
Presentation &
Visualisation
Diagram adapted from The two Crows Corporation: Introduction to Data Mining and Knowledge Discovery. Third Edition
Knowledge Engineering
• What is knowledge engineering?
• What role does it play in the project?
• What do I hope to achieve in the
project?
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Knowledge Engineering
• Definition
– It is the acquisition, validation, representation and explanation
of knowledge.
• Primary activities
– Activities of KE are broad. Only a small subset will be
implemented for the project.
– Knowledge acquisition
• The process of gathering the knowledge to stock the expert
system's knowledge base.
– Knowledge validation
• Objective is to produce knowledge of high integrity
• Validation of knowledge to source and expected or known
outcomes (close collaboration with data miner)????
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Knowledge Engineering
• Role in project
– Information acquisition
– Information preparation (support data
mining)
– Ordered/indexed storage of information
and user profile information
– Support for data mining
• Persistence of data mining deliverables
• Ordered/structured information that can be
efficiently and easily queried
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Knowledge Engineering
• Questions tackled
– Does knowledge engineering effectively
support adaptive knowledge
management?
– Can knowledge engineering further the
functional scope of adaptive knowledge
management?
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Knowledge Engineering
• Success factors
– Responsive support for data mining
– Secure, ordered persistence of
information, user data and data mining
deliverables
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Graphical illustration of roles as they fit into the KDD process
Presentation of
Knowledge to
User
Domain of
Knowledge
Engineering
Domain of
Data Mining
Domain of
Adaptive
Presentation &
Visualisation
Diagram adapted from The two Crows Corporation: Introduction to Data Mining and Knowledge Discovery. Third Edition
Data Mining
• What is data mining?
– Data mining is the step in the knowledge discovery process
that consists of applying data analysis and discovery
algorithms that, under acceptable computational efficiency
limitations, produce a particular enumeration of patterns (or
models) over the data.
• What is the goal of this component?
– Discovery of “Knowledge”
– A subset of this task involves trying to establish criteria for
evaluating the inherent subjective nature of interestingness in
a more objective manner in the given domain of interest.
• 3 Main areas that require mining:
– Concept Mining of documents
– User profiles
– System Learning (unlikely to be implemented)
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Concept Mining
• What is Concept Mining?
– The concept mining area of the system is responsible for
the efficient, timely extraction of knowledge or concepts
from the media stored in the system’s databases.
– Concepts can be interrelated and are related to user
profiles and expected profiles
• How will this be implemented?
– Mainly mining of concepts from unstructured/text documents
– Look at other media including audio, images and video as time
permits
– Evaluating known techniques to select most appropriate
technique
– Component based distributed artificial intelligence
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Profile Mining
• Expected Profile vs User Profile
• What is the expected profile
– The profile that the company expects from a person occupying
a specific job i.e., the knowledge required to render a
satisfactory (and approaching excellence) job performanance
• What is the user profile
– The profile that the user actually has dependant on level of
education, personal interests and knowledge acquired from
company knowledge management system
• How will this be implemented?
– Expected profile from company job descriptions and users tacit
knowledge of job
– User profile mined from usage of the system and personnel
records
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Questions tackled
 How does adapt to its dynamically
changing knowledge in the
enterprise
 How does it incrementally learn new
knowledge
 What is the best AI technology to
use to learn from and adapt to
dynamically changing knowledge
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Success Factors
•
The accuracy and depth of concepts mined from
the existing resources.
•
The most important measure of the success will
be the feedback provided by the users of the
system.
•
How well can adaptive KM force the 2 profiles
to converge for overall sustainability of the
company?
•
How dynamic the learning and adaptive process
is?
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Graphical illustration of roles as they fit into the KDD process
Presentation of
Knowledge to
User
Domain of
Knowledge
Engineering
Domain of
Data Mining
Domain of
Adaptive
Presentation &
Visualisation
Diagram adapted from The two Crows Corporation: Introduction to Data Mining and Knowledge Discovery. Third Edition
Adaptive presentation and
visualization
• Adaptive?
• Presentation?
• Visualization?
• Example:
– If you are a new user and need to find something
amongst hundreds of sources of information you will not
want to look at irrelevant documents, being too detailed
or too broad
– After a while the engine should pick up your patterns
and reduce the data retrieved by the data mining engine
by tailoring it to user needs and expected profile
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Why must the interface be
adaptive?
• Vast amount of data being mined from
different sources
• This will mean that the detail of the data
returned will vary
• It may also be “hidden” – not known to
anybody as the relationships may never
have existed before now
• The data may be complex – many sources
and duplication leads to data being
unusable
• Data returned may be unrelated to search
Henry Brown
Colin Rouse
Phumelelakahle Kunene
What are the problems?
• Users have different needs and use
tools in different ways
• Expected profiles differ from user
profiles
• Must adapt to these different needs
• Use profiles to capture needs
• Profiles are then used to display data
accordingly
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Are there any benefits?
• Knowledge transfer can be sped up
and more effective
• Increases the employee’s value and
so the company’s value
• Less output will be returned
• Thus, the system will be faster
• Increase attention
Henry Brown
Colin Rouse
Phumelelakahle Kunene
Related work
Agents are used to
monitor the user and
profile him
They then act
accordingly and
provide the best
results
Outride now work with
Google to implement
personalized searching
Henry Brown
Colin Rouse
Phumelelakahle Kunene