Westphal

1
Westphal Blaxton Chapter 3 Defining the
Problems to be Solved

• MIS 6473- DATA MINING
• Dr. Segall
• Spring 2004

2
Defining the Problems to be Solved

• Analyzing data requires that you have a broad
  sense of how to classify the types of information
  and the knowledge that you are working with as
  well as how to construct an analysis so as to
  move from one class of knowledge to another.

3
Defining the Problems cont.

• Analyst must be flexible.
• Four frameworks
• Think of knowledge being represented in a
  hierarchy ranging from single objects all the way
  up to full systems
• Distinguish between knowing how procedures are
  accomplished opposed to knowing that certain
  facts are true about the situation
• Characterize your problem in terms of
  metaknowledge and actual knowledge
• Map your problem onto the orthogonal dimensions
  of situations versus parameter values.

4
Defining the Problem cont.

• Once the problem is mapped onto a conceptual
  framework, analyst must think about whether the
  analysis will proceed in the reactive mode,
  proactive mode, or a combination of both.

5
Challenging Analysts to Think Outside the Box

• Analyst becomes comfortable with few, select, and
  limited choices.
• Rat Analogy
• Driving without maps and cooking without recipes.
• No road maps for data mining
• Southern Cooking requires no recipes only simple
  ingredients. Different dishes are a result of
  variation in mixing the simple ingredients.

6
Theres more than one way to slice a Bagel

• How would you cut a bagel in half?
• Find out the real requirements of a task before
  taking any action to complete it.

7
Mapping your problem onto a Hierarchical Framework
8
From Objects to Networks Applications and Systems

• Basic unit of knowledge representation is an
  object
• Relationships are like objects, they are mutually
  exclusive of one another
• Relationships and objects together form networks
  of data
• The integration of one or more networks forms the
  basis of an application
• Applications are usually stand alone solutions
  developed for a particular problem or targeted to
  a specific domain
• Systems are an environment used by analysts to
  perform a wide range of data mining activities

9
Distinguishing between Knowing How and Knowing
That Procedural vs. Declarative Knowledge

• Procedural knowledge knowing how to do things
• Declarative knowledge or knowing that,
  represents factual information about the world.

10
Breaking Declarative Knowledge into Subcategories

• Two types of declarative knowledge episodic or
  semantic.
• Episodic- temporal or spatial information
• Semantic- descriptive representations

11
Distinguishing between Metaknowledge and Actual
Knowledge

• First dimension Actual status
• Second dimension Metaknowledge
• Four categories of knowledge
• YKYK
• YKDK
• DKYK
• DKDK

12
The information that You Know You Know (YKYK)

• This is the simplest case of information
  knowledge.
• It is information that actually exists and the
  user is aware that the information is there.
• Ex Water boils at 100 degrees Celsius.

13
The information that You Know You Dont Know
(YKDK)

• Information that is generally not known or
  readily accessible but can be researched to find
  the answer.
• Ex - Exact height of Empire State Building
• - The area of Central Park
• - How many vehicles are registered at ASU.

14
The information that You Dont Know You Know
(DKYK)

• Type of information targeted in exploratory
  analyses where you may not have definite ideas
  about what is expected to be found.
• Discovery of information that is already
  accessible but not currently being used.
• Ex - exposure of fraud through investigation
  (and a little snooping)

15
The information that You Dont Know You Dont
Know (DKDK)

• The most vulnerable situation and it affects all
  aspects of business.
• The boundaries of the DKDK knowledge are
  undefined and based on unknown parameters.

16
Dimensions of Metaknowledge and Actual knowledge
within the problem space
17
Distinguishing between Situations and Parameter
Values

• We can recast the world into different paradigm.
• Conducting a financial transaction, making a
  telephone call, or fixing a car.
• All have a set of parameters.
• Conduct a matrix

18
Known Situation and Established Parameter
Boundaries

• Risks and probabilities
• Exceptions easily flagged
• Example Going over on credit card.

19
Unknown Situation and Established Parameter
Boundaries

• Bottom-up analyses
• Detect patterns in existing data
• Example - Medical case reports

20
Unknown Situation and No Established Parameter
Boundaries

• Most threatening circumstances
• Proactive invigilation
• Behavior is not yet discovered

21
Figure 3.3
22
Performing Analyses in Reactive and Proactive
Modes

• Reactive Analysis
• Proactive Analysis

23
Performing Reactive Analysis

• Concrete question? yes/no
• Analysis is on entity
• Indexed structures

24
Performing Proactive Analysis

• Unknown and cannot be defined
• Big picture
• Structuring the Proactive Slice of Data

25
Combining Proactive and Reactive Techniques

• Work on terabytes of data
• Combination to navigate through the data
• Example - Internet

26
Figure 3.4
27
Summing Up

• Thinking of data
• Next get the data in shape

28
Dr. Segalls Questions
29
Dr. Segalls Question 1

• An object is the basic unit of knowledge
  representation. At this level, what are all
  patterns and trends based on and explain how this
  relates to the hierarchical framework for
  knowledge.

30
Dr. Segalls Question 2

• What are the four dimensions of actual or
  metaknowledge? Explain each of these.

31
Dr. Segalls Question 3

• Explain what is the bottom up analyses.