A Generic Software Framework for building Hybrid Ontology-Backed Models for Driving Applications

1
A Generic Software Frameworkfor buildingHybrid
Ontology-Backed ModelsforDriving Applications

• Colin Puleston, James Cunningham, Alan Rector
• Bio-Health Informatics Group, University of
  Manchester

2
Hybrid-Model Demo
3
Demo Shows

• Object Model (Java) provides central structure
• OWL Ontology provides detailed domain knowledge
• Dynamic model updates
• Orchestrated by Object Model
• Based on interaction with ontology classifier

4
Object-Model/Ontology Interaction
5
Ontological knowledge comes from set of External
Knowledge Sources (EKS)
Object Model accesses EKS and associated
inference mechanisms via mediating
interface has no knowledge of any EKS
format and no knowledge of any inference
mechanisms
Object Model
EKS Access System
OWL-Access
Protege-Ac.
FaCT
Domain Knowledge Sources (ontologies, databases,
etc.)
6
Ontological knowledge comes from set of External
Knowledge Sources (EKS)
Object Model has no prior knowledge of any EKS
contents other than an expectation that a
small number of key-entities will be
represented somewhere and that mappings will be
provided via a configuration file
Object Model
EKS Access System
OWL-Access
Protege-Ac.
FaCT
Domain Knowledge Sources (ontologies, databases,
etc.)
7
Object Model
field-value updated
Cancer
C1
selected
Get concept-hierarchy for ltproblem-type-identifier
gt
Update field-set constraints for
Create instance of
C1a
C1
Cancer
C1
EKS Access System
8
Concept-Instance state update (OWL
Implementation)
Concept-Instance
Concept-Instance
Cancer
Instance-of
Cancer
Instance-of
Field
Root
Value
locus
Locus
Breast
Field
Root
Value
stage
BCs
Stage2
locus
Locus
Breast
subSt.
BC2ss
BC2ss
stage
Stage2
BCs
Update (via sanctioning mechanism)
To OWL Description
Stage2BreastCancer EquivalentTo
BreastCancer and stage some
Stage2 SubClassOf subStage some (SubStageA
or SubStageB)
Cancer and locus some Breast and stage some Stage2
Classify
9
Note

• EKS Access System allows compositional concepts
  such as BreastCancer and Stage2BreastCancer to be
  hidden from Object Model
• Hence No need for classification during
  execution of queries constructed via Object Model
• Hence Relatively efficient (SPARQL-based)
  querying of large numbers of individual records

10
Why Hybrid Models?
11
For simple GLIMPSE model of time
Cancer
Breast
Stage2
SubStageA
type
locus
subStage
stage
Problem-Glimpse
...simple one-to-one correspondence between
Object Model and EKS Access System
Concept-Instance
Cancer
Instance-of
Field
Root
Value
locus
Locus
Breast
stage
BCs
Stage2
subSt.
BC2ss
BC2ss
12
But full SNAP/SPAN model of time involves
non-ontologically-specifiable interaction
Cancer
Breast
type
locus
Problem-History
snapshot
snapshot
snapshot
Problem-Snapshot
Problem-Snapshot
Problem-Snapshot
stage
stage
subStage
stage
Stage1
Stage1
Stage3
SubStageA
13
Including interaction with a Temporal Abstraction
System
Stage1
Stage3
Stage1
Stage3
min
max
start
end
Temporal-Abstractions
Cancer
Breast
type
locus
stage
Problem-History
snapshot
snapshot
snapshot
Problem-Snapshot
Problem-Snapshot
Problem-Snapshot
stage
stage
subStage
stage
Stage1
Stage1
Stage3
SubStageA
14
Hence, more complex mappings between Object Model
and EKS Access System
Concept-Instance
Cancer
Instance-of
Field
Root
Value
locus
Locus
Breast
stage
BCs
Stage1
Cancer
Breast
type
locus
Problem-History
snapshot
Problem-Snapshot
stage
Stage1
15
Hence, more complex mappings between Object Model
and EKS Access System
Concept-Instance
Cancer
Instance-of
Field
Root
Value
locus
Locus
Breast
stage
BCs
Stage1
Cancer
Breast
type
locus
Problem-History
snapshot
Problem-Snapshot
stage
Stage1
16
Hence, more complex mappings between Object Model
and EKS Access System
Concept-Instance
Cancer
Instance-of
Field
Root
Value
locus
Locus
Breast
stage
BCs
Stage3
subSt.
BC2ss
SSA
Cancer
Breast
type
locus
Problem-History
snapshot
Problem-Snapshot
subStage
stage
Stage3
SubStageA
17
Hence, more complex mappings between Object Model
and EKS Access System
Stage1
Stage3
Stage1
Stage3
min
max
start
end
Temporal-Abstractions
Cancer
Breast
type
locus
stage
Problem-History
Concept-Instance
also involving Temporal Abstraction System
Temporal-Abstractor-Set
Cancer
Instance-of

Field
Root
Value
Field
Type
locus
Locus
Breast
stage
ORDINAL
stage
BCs
???
18
Hybrid Models Summary

• Non-ontologically specifiable interaction
• Hence, hybrid models of some type required
• Advantages of Object-Model based solution
• Relatively simple
• Flexible
• Coherent
• Domain-specific API

19
Higher-Level Model Architecture
20
Higher-level interaction orchestrated by
Interaction Model
Interaction Model
EKS Access System
Temporal Abstraction System
Static Model (SNAP/SPAN)
Object Model
21
Additional Processing Mechanisms distinct from
Interaction Model - do not contribute to shape of
model itself
Interaction Model
EKS Access System
Temporal Abstraction System
Static Model (SNAP/SPAN)
Additional Processing Mechanisms
Object Model
22
Driving Applications
23
Basic Source Representation provided by
collection of domain-specific Java classes
7

• Domain-specific API
• Interaction Model intrinsic to domain classes
  (extending generic patterns where applicable)

V
4.3
W
B
A
E
D
X
C
Z
2
Y
8/1/97
Object of (domain-specific) class A
Relationship represented by object of class Link
Field on object
A
24
Automatic translation into domain-neutral Network
Representation

• Domain-neutral semantic network representation
• Interaction Model translated into a set of
  listeners
• All translation specification is by individual
  classes

7
V
4.3
W
B
A
E
D
X
C
Z
2
Y
8/1/97
Relationship represented by object of class Link
Object of class Node - representing instance of
concept A
A
25
Query formulation via extended version of Network
Representation

• Extensions for
• Insertion of logical operators
• Specification of data-value constraints
  (including temporal constraints)
• Specification of temporal-abstractions over
  dynamically defined time-periods (not shown)

?p
D
gt ?p
A
C
Z
OR
B
OR
Y
NOT
E
X
F
?q
lt (?p ?q) / 2
26
Software
27
Core Model-Builder provides generic
EKS-interaction mechanisms, skeleton
interaction-patterns, etc. Chronicle
Model-Builder provides core SNAP/SPAN
representation, interaction with Temporal
Abstraction System, chronicle-specific
interaction-patterns, etc.
Patient Chronicle Model
Temporal Abstraction System
Chronicle Model-Builder
Core Model-Builder
EKS Access System
OWL-Access
Protege-Ac.
28
GUIs (Domain-Neutral)
Record Browser
Model Browser
Query Formulator
Data-Creation Tools (Domain-Specific)
Patient Chronicle Simulator
Model
Patient Record Chronicliser
Storage Manager
Query Engine
Temporal Abstraction Extension
EHR Store
RDF Store
Data-Storage/Query System (Domain-Neutral)
29
Further demos available on request please
ask!Software soon available on-line(Technical
Supplement will provide details)