Data and Knowledge Representation Lecture 3

1
Data and Knowledge RepresentationLecture 3

• Qing Zeng, Ph.D.

2
Last Time We Talked About

• Boolean Algebra
• Predicate Logic (First order logic)

3
Today We Will Talk About

• Ontology
• Major KR Schemes

4
Tell me whats in this room

• Tables, chairs, windows, computers, papers, pens,
  people, etc..
• We can write
• But what is a table? What is a room?
• Logic has no vocabulary of its own

5
Ontology Fills the Gap

• Ontology is a study of existence, of all kinds of
  existence, of all kinds of entities
• It supplies the predicates of predicate logic and
  labels that fill the boxes and circles of
  conceptual graph

6
Websters Definition of Ontology

• 1 a branch of metaphysics concerned with the
  nature and relations of being2 a particular
  theory about the nature of being or the kinds of
  existents -- http//www.webster.com/cgi-bin/dicti
  onary

7
My Simplified Understanding

• Ontology seeks to describe entities through
  classification of relations among entities
• Domain ontology limits the its scope to a
  specific domain such as medicine
• In informatics, we further limit domain ontology
  to what is needed by a application or certain
  kinds of applications such clinical guideline,
  retrieval of pathology information

8
Why Ontology in Biomedical Domain

• Encode data
• E.g. Patient A is diabetic and HIV positive
• Represent knowledge
• E.g. Blood Glucose test is a diagnostic test for
  diabetes.

9
Sources of Ontology

• Observation provides knowledge of the physical
  world
• Reasoning make sense of observation by
  generating a framework of abstractions called
  metaphysics.

10
Ontology Development in Biomedical Domain

• Areas that directly involve ontology
• Data model
• Vocabulary/terminology
• Knowledge based system

11
Philosophers Approach to Ontology

• Top-down
• Concerned with the entire universe
• Build top level ontology first
• Long history
• Lao Zi (Book of Tao)
• Plato
• Aristotle
• Kant (1787)

12
Computer/Information Sciences Approach

• Bottom Up
• Start with limited world or specific applications
• Exception Cyc system
• Designed with computing in mind
• Short History
• First use of the term ontology in computer
  science community McCarthy, J. 1980
  Circumscription A Form of Non-Monotonic
  Reasoning, Artificial Intelligence, 5 13, 2739.

13
Problems Faced by Computer/Information Scientists

• Tower of Babel
• Ontology used/developed by different groups for
  applications
• Terminological and conceptual incompatibilities
• Problem arise in system development and
  maintenance as well as data/knowledge exchange
• Insufficient expressive power

14
Example

• Problem Oriented Medical Record
• Weed LL. Medical records that guide and teach.
  1968. MD Comput. 1993 Mar-Apr10(2)100-14.
• Where SOAP comes from
• The gist organizing medical data/information by
  patient problem
• Many EMRs has a place for problem list

15
Example

• Which one of the following is a problem
• Cough
• Anxiety
• Pregnancy
• Sleep disorder
• Rash
• Physicians can not agree
• Cited by a number of POEMRs as one of the reasons
  of failure

16
Another Example

• What does acute mean?
• sharpness or severity e.g. acute pain
• having a sudden onset, sharp rise, and short
  course, e.g. acute pancreatitis
• In a data model for finding, we had severity as
  an attribute. Thus need to decide where acute fit
  in.

17
To Solve the Problem

• Develop formalism for sharing (e.g. KIF, CGIF)
• Develop standard ontology
• Develop new formalism to increase expressive power

18
Ontological Categories

• Making a choice on ontological categories is
  first step in system design John Sowa
• Ontological Categories is
• Class in OO system
• Domain in database theory
• type in AI theory
• type or sort in logic

19
Ontological Categories

• Making a choice on ontological categories is
  first step in system design John Sowa
• Ontological Categories is
• Class in OO system
• Domain in database theory
• type in AI theory
• type or sort in logic

20
Brentanos tree of Aristotles Categories
Being
Accident
Substance
Property
Inherence
Relation
Directness
Containment
21
CYC Ontology
Thing
Represented Thing
Individual Object
Intangible
Relationship
Event
Stuff
IntangibleObject
Collection
22
Contrast -gt Distinction

• All perceptions start with contrast
• Bright dark
• Tall short
• Healthy ill
• Happy sad
• Distinction (discrete/continuous) conceptual
  interpretations of perceptual contrasts

23
Contrast -gt Distinction

• All perceptions start with contrast
• Bright dark
• Tall short
• Healthy ill
• Happy sad
• Distinction (discrete/continuous) conceptual
  interpretations of perceptual contrasts

24
Distinction -gt Categories

• Distinctions maybe combined to generate
  categories. E.g.
• Classify patients.
• Distinctions (insured, uninsured), (inpatient,
  outpatient), (infant, child, adult), (emergency,
  urgent, general)..
• Categories insured pediatric emergency patient,
  uninsured adult inpatient

25
Sowas Ontology (Peirce and Whitehead)

• AXIOMS
• Physical physical entities have location in
  space and a point in time. E.g. hand, hair,
  computer.
• Abstract abstract entities do not have location
  in space or a point in time. E.g. theorem,
  knowledge, story.

26
Sowas Ontology

• AXIOMS
• Independent independent entities can exist
  without being dependent on the existence of
  another entity. E.g. person, diary, song.
• Relative relative entities require the existence
  of some other entity. E.g. joints between bones,
  middle child, remission after a disease episode.
• Mediating mediating entities require the
  existence of (at least) two other entities and
  establish new relationship among them. E.g.
  theory of relativity, diagnostic strategy,
  cardiovascular system.

27
Sowas Ontology

• AXIOMS
• Continuant has only spatial parts and no
  temporal parts identity cannot depend on
  location in space and time. E.g. gender, alert
  and reminder system, medication formula.
• Occurrant has both spatial parts (participants)
  and no temporal parts (stages) can only identify
  by location in space and time. E.g. disease
  episode, clinical event, medication order.

28
Matrix of Central Categories
Physical Physical Abstract Abstract
Continuant Occurrent Continuant Occurrent
Indepen-dent Object Process Schema Script
Relative Juncture Participation Description History
Mediating Structure Situation Reason Purpose
29
Exercise

• Assume you are developing an alert system to
  monitor errors in laboratory information systems.
  Identify some distinctions for categorizing the
  errors and describe which distinctions are in
  contrast with which other distinctions.

30
Semantic Network

• An long existing notion there are different
  pieces of knowledge of world, and they are all
  linked together through certain semantics.

31
Basic Components

• Nodes
• Represent concepts
• Arcs
• Represent relations
• Labels for nodes and arcs

32
Little Constraint
patient
Interact
Interact
Nurse
physician
Interact
33
Little Constraint
DSG Site
Link
Link
Instructors Homepage
Course Site
Link
Web
34
Relation

• Directed or non-directed
• Multiple relations between two concepts
• Can have different properties
• Reflexive (e.g. co-ocurrence)
• Transitive (e.g. causal)
• Symmetric (e.g. sibling)
• ..

35
Some Often Used Relations in Biomedical Domain

• IS A
• IS PART OF
• CAUSE OF
• MEASURES
• CO-OCCURS

36
Major Limitation

• Lack of Semantics
• No formal semantic of the relations
• E.g. Does ISA mean subclass, member, etc?
• Possible multiple interpretations
• Restricted expressiveness
• E.g. can not distinguish between instance and
  class

37
Extension

• Extending expressivity (distinguish different
  types of concepts and relations
• Distinguish between some and all
• Distinguish between existence and intension
• Distinguish between definition and assertion
• Add semantic rigor
• Map to logic (Sowa CG)

38
Frame-based Network

• Distinguish instance vs. class
• Hierarchical structure (superclass and subclass)
• Multiple hierarchy
• Slots
• Member slot
• Own slot

39
Slot

• Frame identifying information
• Relationship between frames
• Descriptors of requirements for frame match
• Procedural information
• Default information
• Restrictions and constraints
• New instance information

40
Strength

• Help organize knowledge hierarchically
• Procedure information
• Support multiple inheritance

41
Weakness

• Expressiveness (e.g. quantifier)
• Inheritance
• Sub classing (override slot value)
• Multiple inheritance
• Large complex knowledge system

42
Example MED
43
Example Protégé
44
Example Protégé
45
Example Protégé
46
Production Rules

• Also called IF-THEN rules
• Many forms
• IF condition THEN action
• IF premise THEN conclusion
• IF proposition p1 and proposition p2 are true
  THEN proposition p3 is true

47
Components

• Rule base
• Inference engine
• Working memory

48
Inference

• Modus ponens
• Forward chaining
• Modus tollens
• Background chaining

49
Example MYCIN

• IF the identity of the germ is not known with
  certainty
• AND the germ is gram-positive
• AND the morphology of the organism is "rod"
• AND the germ is aerobic
• THEN there is a strong probability (0.8) that the
  germ is of type enterobacteriacae

50
Example

• POINT

Main Inference Control
Control the execution of inference engine
by retrieve and providing needed knowledge
Jess Inference Engine
Fire rules when adequate knowledge is provided
Medical Knowledge base
Inference Rules
Define semantic relations between concepts
Define rules of relevance base on semantic
relations between concepts
51
Example
Medical Knowledgebase
Inference Rules
Inference Process
Inference Results
52
Pro and Con

• Pro
• Modular
• Natural
• Con
• Not efficient
• Not expressive

53
Exercise

• The thyroid gland is located at the base of your
  neck in front of your trachea (or windpipe). It
  has two sides and is shaped like a butterfly.
• The thyroid gland makes, stores, and releases two
  hormones - T4 (thyroxine) and T3
  (triiodothyronine). Thyroid hormones control the
  rate at which every part of your body works. This
  is called your metabolism. Your metabolism
  controls whether you feel hot or cold or tired or
  rested. When your thyroid gland is working the
  way it should, your metabolism stays at a steady
  pace -not too fast or too slow.
• If no cancer cells are found, your doctor may
  prescribe a thyroid hormone to decrease the size
  of your nodule. Or, your doctor may suggest
  surgery to remove it. If cancer cells are found,
  further treatment will be needed. Thyroid cancer
  usually can be treated with success.

54
Excise

• Which representation scheme to choose?

55
Reading

• Sowa Chapter 2
• Sowa Chapter 4