Web Ontology Segmentation Analysis, Classification and Use Julian - PowerPoint PPT Presentation

Web Ontology Segmentation Analysis, Classification and Use Julian Seidenberg Alan Rector jms@cs.manchester.ac.uk rector@cs.manchester.ac.uk

Ontology ?! Definition: An ontology describes concepts in a domain of interest and the relationships that hold between them. 2

Use Schema mapping Knowledge capture Information inference Intelligent user interfaces and ... 3

global interoperability 4

Social Medicine Biology Science Particle Astronomy Ecology Physics large domain ontologies 5

number of time to classify concepts Gene Ontology 20,000 220 sec. GALEN 23,000 stack overflow! NCI-Thesaurus 42,000 815 sec. FMA 70,000 stack overflow! SNOMED-CT 364,000 stack overflow! (reasoner: Racer 1.7) 6

4000 years ago ...

3 styles of segmentation

partitioning 12

? ? querying 13

? traversal 14

Background: superclasses as links GeneralisedStructure Heart ⊆ isA isA AbstractStructure BodyStructure isA isA InternalOrgan BodySystem InternalOrgan isA isA ∃ isPartOf . CardiovascularSystem Heart CardiovascularSystem ∃ isPartOf 15

Segmentation by Traversal cross-link cross-link cross-link cross-link cross-link Target of Target of Target of Target of Target of Segmentation Segmentation Segmentation Segmentation Segmentation cross-link cross-link cross-link cross-link cross-link 16

Test Case • Big 23,000 classes • Complex 10,000 definitions 30,000 relationships • Representative basis of many future systems 17

25,000 20,000 15,000 10,000 5,000 0 Original Segment number of classes segment of the ‘Heart’ 18

Boundary Limiting 19

Limitation (Boundary Limiting) missing information ↓ lost inferences 20

Good control over number of classes size But does not optimize 6,000 classification 4,500 3,000 1,500 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 20 30 boundary depth boundary limit vs. size 21

Property Filtering (partitive filter) hasPart hasPart hasStatus hasPart hasPart hasLocation hasColor Target of Target of hasLocation Segmentation Segmentation 22

Limitation (Property Filtering) modified definitions ↓ unexpected inferences 23

Definition Transformation SkinOfFrontalScalp ≡ (SkinOfScalp ∩ ∃ hasProximity . FrontalBone) SkinOfFrontalScalp ≡ SkinOfScalp SkinOfFrontalScalp ⊆ SkinOfScalp 24

6,000 4,500 3,000 1,500 0 c e d e e l l r a a i e v o v v s r n i i i i a u m f t t t o i c a i b t d t i u c c t r o + c o u r a m n t l p r c s t u n n s f u o f c + t r a p number of class 25

85 63 seconds 41 19 -3 c e d e e l l r a a i e v o v v s r n i i i i a u m f t t t o i c a i b t d t i u c c t r o + c o u r a m n t l p r c s t u n n s f u o f c + t r a p classification time 26

loading scanning extraction saving Constructive Part + Func + Mod Locative Partitive Structural Functional Modifier 0 15 30 45 60 75 90 105 120 seconds to create segment 27

Summary • 3 different styles of segmentation Partitioning, querying, & traversal • Segmentation by traversal Semi-automatic Works well with densely interconnected ontologies Independently coherent subset (not simply a view) • Boundary limiting Accurately controls size • Property filtering Significantly increases tractability Facilitates ontology profiling 28

Questions ? Julian Seidenberg jms@cs.manchester.ac.uk 29