1 A Conceptual Framework of Data Mining Y.Y. Yao Department of Computer Science, University of Regina Regina, Sask., Canada S4S 0A2 [email protected].

1

A Conceptual Framework of Data Mining

Y.Y. Yao

Department of Computer Science, University of ReginaRegina, Sask., Canada S4S 0A2

[email protected]://www.cs.uregina.ca/~yyao

2

Acknowledgements

Thanks to Professors Wang Jue Zhou Zhi-Hua Zhou Aoying for the kind invitation and this

opportunity.

3

Motivations

“The question typically is not what is an ecosystem, but how do we measure certain relationships between populations, how do some variables correlate with other variables, and how can we use this knowledge to extend our domain.” Salthe, S.N. Evolving Hierarchical Systems, Their Structure and Representation

4

Motivations

“… the scientist is usually not, on the other hand, a self-conscious epistemologist. That would mean going beyond his area of narrow training for the purpose of questioning its point. Functioning as a scientist means functioning within the rules of a game learned during the apprenticeship in which examination of the philosophic foundations of the game plays a characteristically tiny role.”

5

Motivations (Data Mining)

One is more interested in the algorithms for finding “knowledge”, but not what is knowledge.

One is more interested in a more implementation-oriented view or framework of data mining, rather than a conceptual framework for the understanding of the nature of data mining.

6

Data mining Function-oriented approaches:

Requirements Theory-oriented approaches:

Mathematical/statistical methods Procedure/process-oriented approaches:

KDD processes

There does not exist a concept framework for data mining.

7

Motivations (General) We are more interested in doing than

understanding. We are more interested in actual systems and

methods than a powerful point of view. We are more interested in solving a real world

problem than acquisition of knowledge.

We have enough knowledge, but not sufficient wisdom in using the knowledge.

8

Motivations Four international workshops have

been held on foundations of data mining.

There still does not exist a well accepted and non-controversial framework.

Many papers do not cover the “foundations of data mining”.

9

The question

How to view and study data mining?

What can we learn from our experiences?

From other fields. From well established branches.

10

Knowledge structure and problem solving in physics

Reif and Heller, 1982.“Effective problem solving in a realistic domain

depends crucially on the content and structure of the knowledge about the particular domain.”

The knowledge about physics “specifies special descriptive concepts and relations described at various level of abstractness, is organized hierarchically, and is accompanied by explicit guidelines specifying when and how this knowledge is to be applied.”

11

Knowledge structure and education

• Experts and novices differ in their knowledge organization.

• Experts are able to establish multiple representations of the same problem at different levels of granularity.

• Experts are able to see the connections between different grain-sized knowledge.

12

Cognitive Science

Posner, 1989 According to the cognitive science

approach, to learn a new filed is to build appropriate cognitive structures and to learn to perform computations that will transform what is know into what is not yet known.

13

A New View

Data mining as a field of study, rather than simply a collections of algorithms, or a combination of several fields.

The study of data mining may be viewed as a scientific enquiry into the nature of data mining and the scope of data mining methods.

14

Three basic questions

What are the foundations of data mining?

What is the scope of the foundations of data mining?

What are the differences between existing researches and the research on the foundations of data mining?

15

A potential solution

The study of the nature of data mining

The study of the nature of data mining

The study of

data mining

methods

The study of

data mining

methods

The philosophical foundations

The theoretical foundations

The mathematical foundations

The philosophical foundations

The theoretical foundations

The mathematical foundations

The technological foundationsThe technological foundations

16

A conceptual framework A layered framework can be

established. Each layer/level deals with the

problem in different contexts: in mind and in the abstract in machine application.

17

A layered model of Data Mining

Philosophy level Algorithm/technique level Application level

Philosophy layer

Technique layer

Application layer

18

A layered model

Philosophy level: What is knowledge?

The study of knowledge & knowledge discovery in mind and in the abstract.

What is knowledge representation?How to express and communicate knowledge?What is the relationship between knowledge in mind and in real world?How to classify knowledge?How to organize knowledge?

Philosophy layer

Technique layer

Application layer

19

A layered model

Technique level: How to discover knowledge?

The study of knowledge & knowledge discovery in machine.How to code, storage, retrieve knowledge in computer?How to develop an efficient algorithm?How to improve an existing technique?

Philosophy layer

Technique layer

Application layer

20

A layered model

Application level: How to use the discovered knowledge

The study of the applications of discovered Knowledge.

Is the discovered knowledge useful?Is the discovered knowledge meaningful?How to use the knowledge?

Philosophy layer

Technique layer

Application layer

21

A layered model

Philosophy level

The study of knowledge & knowledge discovery in mindand in the abstract.

Technique level

The study of knowledge & knowledge discovery in machine.

Application level

The study of the applications of discoveredKnowledge.

1. The division among the three levels is not a clear cut, and may have overlaps with each other.

2. The inner layers establish a foundation for the outer layers.

3. The outer layers may raise questions for the inner layers.

22

A layered model of KDD

The results from philosophy level will provide guideline and set the stage for the algorithm and application levels.

Philosophical study does not depend on the availability of specific techniques.

Technical study is not constrained by a particular application.

The existence of a type of knowledge in data is unrelated to whether we have an algorithm to extract it.

The existence of an algorithm does not necessarily imply that the discovered knowledge is meaningful and useful

23

A layered model of KDD

The three levels represent the understanding, discovery, and utilization of knowledge.

Any of them is indispensable in the study of intelligence and intelligent systems.

They must be considered together in a common framework through multi-disciplinary studies, rather than in isolation.

24

Application of the layered framework Concept formation and learning can

be studied within the layered framework.

The reconsideration brings a better understanding of the problem.

25

Application of the layered framework Concept formation and learning can

be studied within the layered framework.

The reconsideration brings a better understanding of the problem.

26

Philosophy level study of concept

Classical viewA concept is described jointly by its intension and extension.

In ten sion E x ten s ion

C o nce p t

27


28


Two basic issues of concept formation

Aggregation aims at the identification of a group of objects so that they form the extension of a concept.

Characterization attempts to describe a set of objects as their intension.

29


Classical view

DifferentiationDifferentiation

IntegrationIntegration

AggregationAggregation

CharacterizationCharacterization

ConceptformationConcept

formation

30


Classical view



vs.

Differences


formation

31


Classical view



vs.

Differences

Similarities


formation

32


Classical view



vs.

Extension

Intension


formation

33


ContextContext

HierarchyHierarchy

ConceptlearningConceptlearning


formation

34


ContextContext

HierarchyHierarchy

ConceptlearningConceptlearning


formation

35

Technique level study of concept

Search for the intensionSearch for the intension

Given a context -Given a context -

Search for the extensionSearch for the extension

Analyze the concepts relationshipAnalyze the concepts relationship

36


Intensions of concepts defined by a language

37


Intensions of concepts defined by a language

38


Conjunctive concept space

39


Conjunctive concept space

40


41


Extensions of concepts defined by an information table

42


43

Technique level study of concept Extensions of concepts defined by an information table

44

Technique level study of concept Relationship between concepts in an information table

45

Technique level study of concept Relationship between concepts in an information table

46

Technique level study of concept Probabilistic measures:

47

Technique level study of concept Probabilistic measures:

48


Concept learning as searchConcept learning as search

49



50



51



52

Application level study of concept

The main purposes of science are to describe and predict, to improve or manipulate the world around us, to explain our world.

Concepts learning should serve the same purposes.

53


to describe

54


to predict

55

Application level study Domain specific The usefulness of concepts needs to

be defined and interpreted based on other more familiar notions.

56

Conclusions It is important to treat data mining

as a field of scientific enquiry. One needs to consider all aspects of

data mining. The layered framework may provide

a better understanding of data mining.

57

Conclusions We need to find the cognitive

structures or knowledge structures of data mining.

We need to move beyond algorithm and application centered views of data mining.

We need to avoid seductive semantics.

58

Conclusions Data mining can be studied in the

context of scientific discovery and research methods.

Data mining and machine learning systems may be viewed as support systems for the exploration of data, such as research support systems.

59

Thank you!Thank you!

The ideas are preliminary and need fine tune.

You comments, suggestions, and criticisms are welcome!

1 A Conceptual Framework of Data Mining Y.Y. Yao Department of Computer Science, University of Regina Regina, Sask., Canada S4S 0A2 [email protected].

Documents

knowledge structure

acquisition of knowledge

knowledge organization

foundations of data

study data mining

nature of data mining

motivations data miningone

new viewdata mining