TwoLevel Semantic Annotation Model

1
Two-Level Semantic Annotation Model

• BYU Spring Conference 2007
• Yihong Ding

Sponsored by NSF
2
Semantic Web

• Content is represented in
• explicit,
• shared,
• conceptualizations.

• Consequence
• Machines can understand meaning
• Machines can derive implicit logics

3
Semantic Annotation Enable the Semantic Web

• Metadata links data in a web page to defined
  concepts in an ontology
• Annotated data becomes machine-processable

4
Sample Annotation
ltrdfDescription rdfaboutwebpageCarIns13gt lt
carMileagegt116000lt/carMileagegt ltcarPrice"gt3450
lt/carPricegt ltcarMake"gtFORDlt/carMakegt ltcarYea
r"gt1986lt/carYeargt ..... lt/rdfDescriptiongt
5
Two Contradictive Annotation Methods

• Pure ontology-based annotation
• Layout-driven annotation

6
Ontology-based Annotation

• Using defined data-recognition patterns in
  ontologies to directly annotate web content
• Examples in extraction ontologies
• External Representation
• Price \d\d?\d?\d.\d\d
• Contextual Representation
• Context phrases (left, right), e.g. \?
• Context keywords e.g. price obo
  neg(\.otiable)

7
Ontology-based Annotation

• Pros
• Resilient to varying domains and page layouts
• No mapping requirements between recognition
  patterns and ontology definitions
• Cons
• Accuracy depends on how good the knowledge base
  is
• Creation of recognition patterns
• Execution speed

8
Layout-Driven Annotation
lttrgt lttd colspan"2" class"pageHeading"
valign"top"gt lth1gtYou've Got Maillt/h1gt lt/tdgt lt/tr
gt lttrgt lttd align"center" valign"top"
class"smallText" rowspan"2"gt ltscript
language"javascript" type"text/javascript"gt lt!--
document.write('lta href"javascriptpopupWindow(\
'http//www.screen-scraper.com/shop/index.php?main
_pagepopup_imageamppID7\')"gtltimg
src"images/dvd/youve_got_mail.gif" border"0"
alt"You\'ve Got Mail" title" You\'ve Got Mail "
width"100" height"80" hspace"5" vspace"5"
/gtltbr /gtlarger imagelt\/agt') //--gt lt/scriptgt
ltnoscriptgtlta href"http//www.screen-scraper.com
/shop/index.php?main_pageimages/dvd/youve_got_mai
l.gif" target"_blank"gtltimg src"images/dvd/youve_
got_mail.gif" border"0" alt"You've Got Mail"
title" You've Got Mail " width"100" height"80"
hspace"5" vspace"5" /gtltbr /gt larger
imagelt/agtlt/noscriptgt lt/tdgt lttd class"main"
align"center" valign"top"gtModel
DVD-YGEMlt/tdgt lt/trgt lttrgt lttd class"main"
align"center"gtlt/tdgt lt/trgt lttrgt lttd
align"center" class"pageHeadinggt34.99lt/tdgt lttd
class"main" align"center"gtShipping Weight 7.00
lbs.lt/tdgt lt/trgt lttrgt lttdgtnbsplt/tdgt lttd
class"main" align"center"gt10 Units in
Stocklt/tdgt lt/trgt lttrgt lttd class"main"
align"center"gtManufactured by Warnerlt/tdgt lttd
align"center"gt lttable border"0" width"150px"
cellspacing"2" cellpadding"2"gt lttrgt lttd
align"center" class"cartBox"gtnbspQuantity

• Using defined page layout patterns to annotate
  web content
• Example (screen-scraper)

9
Layout-Driven Annotation

• Pros
• More accurate
• Faster
• No knowledge base required
• Cons
• Layout-pattern generation
• Layout-pattern regeneration
• Layout-pattern maintenance
• Mappings between layout-patterns and ontology
  definitions

10
Observation

• These two annotation methods are complementary.

11
Two-Layer Annotation Model
Massive Annotation Process
Structural Annotator
Document
Sample Annotation Process
Conceptual Annotator using ontology-based IE
wrapper
12
lttrgt lttd colspan"2" class"pageHeading"
valign"top"gt lth1gtYou've Got Maillt/h1gt lt/tdgt lt/tr
gt lttrgt lttd align"center" valign"top"
class"smallText" rowspan"2"gt ltscript
language"javascript" type"text/javascript"gt lt!--
document.write('lta href"javascriptpopupWindow(\
'http//www.screen-scraper.com/shop/index.php?main
_pagepopup_imageamppID7\')"gtltimg
src"images/dvd/youve_got_mail.gif" border"0"
alt"You\'ve Got Mail" title" You\'ve Got Mail "
width"100" height"80" hspace"5" vspace"5"
/gtltbr /gtlarger imagelt\/agt') //--gt lt/scriptgt
ltnoscriptgtlta href"http//www.screen-scraper.com
/shop/index.php?main_pageimages/dvd/youve_got_mai
l.gif" target"_blank"gtltimg src"images/dvd/youve_
got_mail.gif" border"0" alt"You've Got Mail"
title" You've Got Mail " width"100" height"80"
hspace"5" vspace"5" /gtltbr /gt larger
imagelt/agtlt/noscriptgt lt/tdgt lttd class"main"
align"center" valign"top"gtModel
DVD-YGEMlt/tdgt lt/trgt lttrgt lttd class"main"
align"center"gtlt/tdgt lt/trgt lttrgt lttd
align"center" class"pageHeadinggt34.99lt/tdgt lttd
class"main" align"center"gtShipping Weight 7.00
lbs.lt/tdgt lt/trgt lttrgt lttdgtnbsplt/tdgt lttd
class"main" align"center"gt10 Units in
Stocklt/tdgt lt/trgt lttrgt lttd class"main"
align"center"gtManufactured by Warnerlt/tdgt lttd
align"center"gt lttable border"0" width"150px"
cellspacing"2" cellpadding"2"gt lttrgt lttd
align"center" class"cartBox"gtnbspQuantity
Data-recognition Pattern (Price) Price
\?\d\d?\d?\d.\d\d
13
Two-Layer Annotation Model, What do we gain?

• All the benefits of the two annotation methods
• Resilient
• Mapping problem eliminated
• Knowledge-base problems mitigated
• Maintenance problem mitigated
• More accurate
• Faster
• Even more
• Automatic augmentation of knowledge base
• Key to large-scale annotation for the web

14
Conclusion

• Contradictory is a synonym of complementary
• Two-level annotation model solves a traditional
  problem on semantic annotation and it is a key to
  large-scale annotation for the web