Transcript Developing Semantic Web Sites: Results and Lessons Learnt

Developing Semantic Web Sites:
Results and Lessons Learnt
Enrico Motta, Yuangui Lei, Martin Dzbor,
Vanessa Lopez, John Domingue, Jianhan
Zhu, Liliana Cabral, Alex Goncalves,
Victoria Uren
Motivation for KMi Sem Web

Key Objective


To generate a live, declarative representation of
what happens in KMi, which can support smart
queries and the specification of intelligent
services producing smart inferences on the
basis of this data
Initial version was ready in 1998

PlanetOnto System (95-98)
Story
Relates-event
Event
People Project Organization Technology
Architecture of Planet-Onto
Query
Interface
Planet KB
NewsBoy
NewsHound
KA Tool
Planet Ontology
Modelling Language (OCML)
WebOnto
Web Browser
Story Database
Email
Architecture of Planet-Onto
Query
Interface
Planet KB
NewsBoy
NewsHound
KA Tool
Planet Ontology
Modelling Language (OCML)
WebOnto
Web Browser
Story Database
Email
Key Criteria for Sem Web Site

Emphasis on Automatic KA

Fully automated generation of information





Manual annotation is welcome but should not be a
core part of the process
Manual annotation should not require sophisticated
KR skills
Ideally manual annotation should take place through
side effects generating from normal work activities
Architecture


No knowledge capture bottleneck
Keep the semantic layer separated (and to some
extent independent) from the actual web site
Interoperability

Semantic Web Site ought to be open

Semantic representation publicly available to any
reasoning engine who wants to use the information
KMi Semantic Web Site
Source Data
Docs
DBs
Integration Layer
Information
Extraction Engine
(Espotter)
Verification Layer
Target Data
Raw
KB
XML mark-up
Data Verification
Engine
DBs
Mapping
Engine
Mapping
Specs
Domain ontology
KB
Ontological Structure
Key Categories
AKT Support Ontology
Publications
AKT Reference
Ontology
Projects
Research Areas
AKT Portal Ontology
People
Organizations
KMi Ontology
Technologies
News
KMi Semantic Web
Data verification


Finding and eliminating duplicate data
Recognizing ambiguous data, e.g. finding
correct person instances for names like
John, Victoria


Using a lexicon component to record the
mappings between strings and instance names
found in the previous processes
Using contextual information to decide
Initial Evaluation
Recall
Number
People
Organizations
Projects
Research
Area
Total
Manual data
93
75
25
23
216
ESpotter finds
77
58
17
13
165
ESpotter Recall-rate
0.827
0.773
0.68
0.565
0.763
Precision
People
Organizatio
ns
Projects
Research Area
Total
Total (discovered)
84
97
19
15
215
Wrong
4
18
1
2
25
Precision rate
0.9523
0.71
0.947
0.86
0.883
KMi Semantic Web
So What?

At a basic level, the architecture works


Services still limited



Developing interesting services requires non
trivial effort
Brittleness is a problem


Automatic generation is key
You rapidly reach the boundaries of the
knowledge held in KMi resources and
performance decreases
Badly needs integration with other similar
resources
No API. Data available only as sources
What should happen next

Integration with other similar activities






Hence this workshop….
Ability to bring in knowledge expressed in
other ontologies
Need for standardised APIs/knowledge
servers
Develop mechanisms for semantic
annotation by side-effect
Improve text mining technology to
improve both the quantity and the quality
of the knowledge
Develop more value-adding services
Intg. with Sem Web Services
Services defined for a particular
Class in a particular Ontology are
available to any system who
asks for them