Representations of Health Concepts:

1
Representations of Health Concepts

• HI6001

2
What is A Representation?
3
(No Transcript)
4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
Representation Determinism

• The format and structure of a representation can
  guide, constrain, and to some extent determine
  the way the mind functions.
• If the structure of a classification system
  matches the users cognitive properties, it can
  increase the users performance. But if there is
  a mismatch, it will decrease the users
  performance.

9
Represented World
The World (Medical entities, objects, phenomena,
events, procedures, etc.)
Vocabularies
Mental Concepts
?
ICD SNOMED UMLS READ CPT-4 NIC
NOC NANDA DICOM NCPDP MeSH etc.
Classical Probabilistic Exemplar Theory-Based LSA
Hierarchical Model Semantic Net
Schemas/Frames Connectionist Construction-
Integration ACT/ACT-R SOAR
Internal Representing World
External Representing World
10
Mental Concepts

• A mental concept is the mental representation of
  a category, which is a set of entities.
• The process of forming a mental concept is called
  categorization.

11
Functions of Mental Concepts

• Classification.
• Understanding and explanation.
• Reasoning.
• Organization.
• Communication.

12
(No Transcript)
13
Theories for Intra-Level Concepts
Theories for Inter-Level Concepts

• Classical View
• Probabilistic View
• Exemplar View
• Theory-Based View
• Latent Semantic Analysis

• Hierarchical Model
• Semantic Network
• Schemas/Frames
• Neural Network
• Construction-Integration
• ACT/ACT-R
• SOAR

14
Three Empirical Effects

• Typicality Effect
• Is robin a bird? Faster
• Is chicken a bird? Slower
• Reversal Effect
• Is chicken a bird? Slower
• Is chicken an animal? Faster
• Relatedness Effect
• Is bat a bird? Slower
• Is bear a bird? Faster

15
Theories for Intra-Level Concepts
16
Classical View

• A set of necessary and sufficient features.
• Features are nested in subset relations.
• Representation of an entire class, not a set of
  exemplars.
• Cannot represent disjunctive concepts.
• (red, square) or (blue, diamond) has no
  necessary features.
• (red, square, large) or (blue, square, small)
  has no jointly sufficient features.

17
Probabilistic View

• Features of a concept have a substantial
  probability of occurring in instances of the
  concept.
• If instance X has some critical sum of weighted
  features of concept Y, then X is a Y.

18
Probabilistic View
19
Probabilistic View

• Concepts are organized around prototypes.
• Prototypes
• A prototype is the best example of a category
• It possesses all characteristic features of a
  category
• It is the central tendency and average of the
  concept.
• Cannot explain context-dependent concepts.
• A harmonic is a typical musical instrument for a
  campfire but not a typical musical instrument for
  a concert.

20
Exemplar View

• The representation of a concept consists of
  separate descriptions of some of its exemplars
  (instances or subsets).

21
Exemplar View
22
Exemplar View

• Classification is based on similarity to a
  particular exemplar.
• A new instance elicits similar old examples.
• Similar instances belong to the same category.
• Conservative with respect to discarding
  potentially relevant information.
• Cannot explain how concepts are created in the
  first place

23
Theory-Based View

• The organization of concepts is knowledge-based
  and theory-driven.
• Categorization is an inference process, not a
  similarity judgment.
• children, money, photo albums, and pets belong
  to a concept, which is things to take out of
  ones house in case of a fire.
• It addresses the question of why we have the
  concepts we have.
• It provides a natural way in which concepts may
  change, that is, through the addition of new
  knowledge and theoretical principles.

24
Theories for Inter-Level Concepts
25
Hierarchical Model
Living Thing
Is-a
Is-a
Plant
Animal
Is-a
Is-a
Bird
Mammal
Is-a
Is-a
Is-a
Is-a
Is-a
Chicken
Cat
Bat
Lion
Robin
26
Hierarchical Model

• Cannot explain typicality effect.
• Cannot explain reversal effect.
• Cannot explain relatedness effect.

27
Semantic Network with Spreading Activation
Animal
Is-a
Is-a
Bird
Is-a
Mammal
Is-a
Is-not-a
Is-a
Is-a
Is-a
has
Is-a
has
Chicken
has
Cat
Bat
Robin
Lion
wings
can
has-a
can
flies
cannot
Red breast
28
Semantic Network with Spreading Activation

• Can explain typicality effect.
• Can explain reversal effect.
• Can explain relatedness effect.

29
Connectionist/Neural Network View

• Concepts are not represented in discrete units
  such as features and instances.
• Concepts are represented as activation patterns
  in parallel and distributed networks of
  neuron-like units that connect to each other and
  excite or inhibit each other through dynamic
  processes.

30
book
oven
bathtub
chair
desk
large
ceiling
floor
sofa
phone
sink
31

• A dynamic updating process maximizes Goodness.
• It is a constraint satisfaction process.

32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
Conclusion 1 Representations of controlled
vocabularies are different from representations
of mental concepts.
36
Represented World
Conclusion 2 Designers of classification systems
should consider theories of mental concepts and
cognitive properties of users.
The World (Medical entities, objects, phenomena,
events, procedures, etc.)
Vocabularies
Mental Concepts
ICD SNOMED UMLS READ CPT-4 NIC
NOC NANDA DICOM NCPDP MeSH etc.
Classical Probabilistic Exemplar Theory-Based
Hierarchical Model Semantic Net
Schemas/Frames Connectionist ACT SOAR
Internal Representing World
External Representing World
37
Conclusion 3 Medical terminology should be
studied and developed as a distributed system
with controlled vocabularies and mental concepts
as two indispensable components.
Distributed Representation System
Vocabularies
Mental Concepts
ICD SNOMED UMLS READ CPT-4 NIC
NOC NANDA DICOM NCPDP MeSH etc.
Classical Probabilistic Exemplar Theory-Based
Hierarchical Model Semantic Net
Schemas/Frames Connectionist ACT SOAR
External Representing World
Internal Representing World