Iterative Ensemble Classification for Relational Data A Case Study - - PowerPoint PPT Presentation

▶
iterative ensemble classification for relational data
SMART_READER_LITE
LIVE PREVIEW

Iterative Ensemble Classification for Relational Data A Case Study - - PowerPoint PPT Presentation

Apr 20, 2023 317 likes •843 views

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Ensemble Classification for Relational Data A Case Study of Semantic Web Services Andreas He and Nicholas Kushmerick University College Dublin, Ireland September

slide-1
SLIDE 1

Outline Introduction Iterative Ensemble Classification Conclusion

Iterative Ensemble Classification for Relational Data

A Case Study of Semantic Web Services Andreas Heß and Nicholas Kushmerick University College Dublin, Ireland September 18, 2004

4 ,

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-2
SLIDE 2

Outline Introduction Iterative Ensemble Classification Conclusion

Outline

1

Introduction Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

2

Iterative Ensemble Classification Iterative Algorithms Specialised Classifiers Evaluation

3

Conclusion Summary Current and Future Work Discussion

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-3
SLIDE 3

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Outline

1

Introduction Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

2

Iterative Ensemble Classification Iterative Algorithms Specialised Classifiers Evaluation

3

Conclusion Summary Current and Future Work Discussion

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-4
SLIDE 4

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Relational Learning

Relational Data consists of objects and relations between objects can be represented as a graphs Three Types of Learning Tasks (following Slattery) classify nodes classify graphs classify subgraphs

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-5
SLIDE 5

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Relational Learning

Relational Data consists of objects and relations between objects can be represented as a graphs Three Types of Learning Tasks (following Slattery) classify nodes classify graphs classify subgraphs

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-6
SLIDE 6

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Classifying Nodes

Task Learn labels for nodes

Instances

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-7
SLIDE 7

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Classifying Subgraphs

Task Learn labels for subgraphs

Instances

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-8
SLIDE 8

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Classifying Graphs

Task Learn labels for graphs

Instances

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-9
SLIDE 9

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Relational Learning: Examples and Methods

Examples for Relational Learning Classical task: classify web pages Methods for relational learning Iterative algorithms Statistical methods

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-10
SLIDE 10

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Two Views

Intrinsic View Features inherent to instance e.g. text from web page Extrinsic View Relations between instances e.g. class labels of linked web pages (Following Neville and Jensen)

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-11
SLIDE 11

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Two Views

Intrinsic View Features inherent to instance e.g. text from web page Extrinsic View Relations between instances e.g. class labels of linked web pages (Following Neville and Jensen)

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-12
SLIDE 12

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Two Views

Intrinsic View Features inherent to instance e.g. text from web page Extrinsic View Relations between instances e.g. class labels of linked web pages (Following Neville and Jensen)

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-13
SLIDE 13

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Two Views

Intrinsic View Features inherent to instance e.g. text from web page Extrinsic View Relations between instances e.g. class labels of linked web pages (Following Neville and Jensen)

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-14
SLIDE 14

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Now for Something Completely Different

1

Introduction Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

2

Iterative Ensemble Classification Iterative Algorithms Specialised Classifiers Evaluation

3

Conclusion Summary Current and Future Work Discussion

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-15
SLIDE 15

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Web Services

Web Services Web-accessible software XML (SOAP) over HTTP Just RPC? Forms? Data Integration?

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-16
SLIDE 16

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Web Service Descriptions (WSDL)

Web Services consist of: Operations (methods) Messages (parameters) Complex types (structures)

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-17
SLIDE 17

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Semantic Web Services

Desired Features Automatic discovery Automatic composition Automatic invocation

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-18
SLIDE 18

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Simple Scenario

Congo WindingStair Teatime

  • author
  • title
  • quantity
  • authName
  • bookT
  • ISBN
  • region
  • qlty
  • qty

? ?

Scenario: Buying a book

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-19
SLIDE 19

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Simple Scenario

Global Ontology

  • Item

➢ Quantity ➢ Price

  • Book

➢ Author ➢ Title ➢ ISBN

  • Tea

➢ Region ➢ Quality

Congo WindingStair Teatime

  • author
  • title
  • quantity
  • authName
  • bookT
  • ISBN
  • region
  • qlty
  • qty

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-20
SLIDE 20

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Simple Scenario

Semantic Metadata (handcrafted)

Global Ontology

  • Item

➢ Quantity ➢ Price

  • Book

➢ Author ➢ Title ➢ ISBN

  • Tea

➢ Region ➢ Quality

Congo WindingStair Teatime

  • author
  • title
  • quantity
  • authName
  • bookT
  • ISBN
  • region
  • qlty
  • qty

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-21
SLIDE 21

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Simple Scenario

Congo WindingStair Teatime

  • author
  • title
  • quantity
  • authName
  • bookT
  • ISBN
  • region
  • qlty
  • qty

!

Global Ontology

  • Item

➢ Quantity ➢ Price

  • Book

➢ Author ➢ Title ➢ ISBN

  • Tea

➢ Region ➢ Quality

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-22
SLIDE 22

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Semantic Metadata

Assumptions Semantic annotation Shared ontology Problem Hand-crafting annotations can be tedious Our Solution Learn mappings from text to ontology

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-23
SLIDE 23

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Semantic Metadata

Assumptions Semantic annotation Shared ontology Problem Hand-crafting annotations can be tedious Our Solution Learn mappings from text to ontology

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-24
SLIDE 24

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Semantic Metadata

Assumptions Semantic annotation Shared ontology Problem Hand-crafting annotations can be tedious Our Solution Learn mappings from text to ontology

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-25
SLIDE 25

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Outline

1

Introduction Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

2

Iterative Ensemble Classification Iterative Algorithms Specialised Classifiers Evaluation

3

Conclusion Summary Current and Future Work Discussion

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-26
SLIDE 26

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Recall: Classifying Nodes

Task Learn labels for nodes

Instances

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-27
SLIDE 27

Outline Introduction Iterative Ensemble Classification Conclusion Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

Mapping Web Services to Relational Learning

Operations Parameters Web Service

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-28
SLIDE 28

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Outline

1

Introduction Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

2

Iterative Ensemble Classification Iterative Algorithms Specialised Classifiers Evaluation

3

Conclusion Summary Current and Future Work Discussion

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-29
SLIDE 29

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Iterative Algorithm for Relational Classification

Iterative Algorithm Features are combination

  • f intrinsic and extrinsic

Extrinsic view changes with new results Predictions are fed back Extrinsic view is dynamic

intrinsic predictions extrinsic/dynamic intrinsic predictions

e.g. Neville/Jensen, Chakrabarti, Lu/Getoor

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-30
SLIDE 30

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Iterative Ensemble Algorithm for Relational Classification

Iterative Ensemble Algorithm Separate intrinsic/extrinsic classifiers Better performance with high-dimensional features

intrinsic extrinsic/dynamic intrinsic predictions predictions

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-31
SLIDE 31

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Iterative Ensemble Algorithm with Specialised Classifiers

Specialised Classifiers Train several classifiers on intrinsic features Each one trained on subset of instances Classifier selected based

  • n extrinsic features

intrinsic intrinsic predictions predictions 2 2 select

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-32
SLIDE 32

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Specialised Classifiers: Example

intrinsic intrinsic predictions 2 select predictions

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-33
SLIDE 33

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Specialised Classifiers: Example

intrinsic intrinsic predictions 2 select

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-34
SLIDE 34

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Specialised Classifiers

Specialised Classifiers Useful if extrinsic view not sufficient on its own Extrinsic features serve as selector Idea can be applied in various ways e.g. Aidan Finn: Information Extraction (16:30h)

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-35
SLIDE 35

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Specialised Classifiers

Specialised Classifiers Useful if extrinsic view not sufficient on its own Extrinsic features serve as selector Idea can be applied in various ways e.g. Aidan Finn: Information Extraction (16:30h)

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-36
SLIDE 36

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Specialised Classifiers

Specialised Classifiers Useful if extrinsic view not sufficient on its own Extrinsic features serve as selector Idea can be applied in various ways e.g. Aidan Finn: Information Extraction (16:30h)

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-37
SLIDE 37

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Specialised Classifiers

Specialised Classifiers Useful if extrinsic view not sufficient on its own Extrinsic features serve as selector Idea can be applied in various ways e.g. Aidan Finn: Information Extraction (16:30h)

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-38
SLIDE 38

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Dataset

Our Web Services Dataset 164 Web Services in 22 Categories 1138 Operations in 136 Domains 5452 Parameters with 312 Datatypes

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-39
SLIDE 39

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Setup

Experimental Setup Using iterative ensemble for service, operations Using specialised classifiers for datatypes Leave-one-service-out Compared to omniscient setup with extrinsic view always correct Compared to non-relational baseline Non-ensemble setting ommitted: always worse than baseline

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-40
SLIDE 40

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Evaluation: Datatype of Parameters

20 30 40 50 60 70 1 2 3 Accuracy % Tolerance Omniscient Specialised Baseline

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-41
SLIDE 41

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Why Allow Tolerance?

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-42
SLIDE 42

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Evaluation: Domain of Operations

20 30 40 50 60 1 2 3 Accuracy % Tolerance Omniscient Iterative Ensemble Baseline

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-43
SLIDE 43

Outline Introduction Iterative Ensemble Classification Conclusion Iterative Algorithms Specialised Classifiers Evaluation

Evaluation: Category of Service

60 70 80 90 1 2 3 Accuracy % Tolerance Omniscient Ierative Ensemble Baseline

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-44
SLIDE 44

Outline Introduction Iterative Ensemble Classification Conclusion Summary Current and Future Work Discussion

Outline

1

Introduction Relational Learning (simplified) Motivation: Semantic Web Services Relational Learning for Web Services

2

Iterative Ensemble Classification Iterative Algorithms Specialised Classifiers Evaluation

3

Conclusion Summary Current and Future Work Discussion

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-45
SLIDE 45

Outline Introduction Iterative Ensemble Classification Conclusion Summary Current and Future Work Discussion

Summary

We have presented... An iterative ensemble method for relational learning Specialised classifiers for relational learning Evaluation has shown... If features high-dimensional: ensemble better than single classifier If extrinsic view alone not sufficient: specialised classifiers

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-46
SLIDE 46

Outline Introduction Iterative Ensemble Classification Conclusion Summary Current and Future Work Discussion

Summary

We have presented... An iterative ensemble method for relational learning Specialised classifiers for relational learning Evaluation has shown... If features high-dimensional: ensemble better than single classifier If extrinsic view alone not sufficient: specialised classifiers

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-47
SLIDE 47

Outline Introduction Iterative Ensemble Classification Conclusion Summary Current and Future Work Discussion

Current and Future Work

Current and Future Work Post-processing result based on mined rules Improve the annotator application Eddie Johnston: Aggregating Web Service output See also our ISWC paper

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-48
SLIDE 48

Outline Introduction Iterative Ensemble Classification Conclusion Summary Current and Future Work Discussion

Current and Future Work

Current and Future Work Post-processing result based on mined rules Improve the annotator application Eddie Johnston: Aggregating Web Service output See also our ISWC paper

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-49
SLIDE 49

Outline Introduction Iterative Ensemble Classification Conclusion Summary Current and Future Work Discussion

Current and Future Work

Current and Future Work Post-processing result based on mined rules Improve the annotator application Eddie Johnston: Aggregating Web Service output See also our ISWC paper

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-50
SLIDE 50

Outline Introduction Iterative Ensemble Classification Conclusion Summary Current and Future Work Discussion

Current and Future Work

Current and Future Work Post-processing result based on mined rules Improve the annotator application Eddie Johnston: Aggregating Web Service output See also our ISWC paper

Andreas Heß Iterative Ensemble Classification for Relational Data

slide-51
SLIDE 51

Outline Introduction Iterative Ensemble Classification Conclusion Summary Current and Future Work Discussion

Discussion

Thank You for Your Attention Questions? URLs and email addresses {andreas.hess,nick}@ucd.ie http://moguntia.ucd.ie, http://www.cs.ucd.ie/staff/nick/ http://smi.ucd.ie/RSWS

Andreas Heß Iterative Ensemble Classification for Relational Data