Finding knowledge, data and answers on the Semantic Web

1
Finding knowledge, data and answers on the
Semantic Web

• Tim Finin
• University of Maryland, Baltimore County
• http//ebiquity.umbc.edu/resource/html/id/202/
• Joint work with Li Ding, Anupam Joshi, Yun Peng,
  Cynthia Parr, Pranam Kolari, Pavan Reddivari,
  Sandor Dornbush, Rong Pan, Akshay Java, Joel
  Sachs, Scott Cost and Vishal Doshi

? http//creativecommons.org/licenses/by-nc-sa/2.0
/ This work was partially supported by DARPA
contract F30602-97-1-0215, NSF grants CCR007080
and IIS9875433 and grants from IBM, Fujitsu and
HP.
2
This talk

• Motivation
• Swoogle Semantic Websearch engine
• Use cases and applications
• Observations
• Conclusions

3
Google has made us smarter
4
But what about our agents?

• Agents still have a very minimal understanding of
  text and images.

5
But what about our agents?

• A Google for knowledge on the Semantic Web is
  needed by software agents and programs

6
This talk

• Motivation
• Swoogle Semantic Websearch engine
• Use cases and applications
• Observations
• Conclusions

7

• http//swoogle.umbc.edu/
• Running since summer 2004
• 1.8M RDF docs, 320M triples, 10K ontologies,15K
  namespaces, 1.3M classes, 175K properties, 43M
  instances, 600 registered users

8
Swoogle Architecture
9
A Hybrid Harvesting Framework
true
Swoogle Sample Dataset
Manual submission
Inductive learner
would
Seeds R
Seeds M
Seeds H
RDF crawling
Bounded HTML crawling
Meta crawling
google
Google API call
crawl
crawl
the Web
10
Performance Site Coverage

• SW06MAR - Basic statistics (Mar 31, 2006)
• 1.3M SWDs from 157K websites
• 268M triples
• 61K SWOs including gt10K in high quality
• 1.4M SWTs using 12K namespaces
• Significance
• Compare with existing works ( DAML crawler,
  scutter )
• Compare SW06MAR with Googles estimated SWDs

SWDs per website
Website
11
Performance crawlers contribution

• High SWD ratio 42 URLs are confirmed as SWD
• Consistent growth rate 3000 SWDs per day
• RDF crawler best harvesting method
• HTML crawler best accuracy
• Meta crawler best in detecting websites

of documents
12
This talk

• Motivation
• Swoogle Semantic Websearch engine
• Use cases and applications
• Observations
• Conclusions

13
Applications and use cases

• Supporting Semantic Web developers
• Ontology designers, vocabulary discovery, whos
  using my ontologies or data?, use analysis,
  errors, statistics, etc.
• Searching specialized collections
• Spire aggregating observations and data from
  biologists
• InferenceWeb searching over and enhancing proofs
• SemNews Text Meaning of news stories
• Supporting SW tools
• Triple shop finding data for SPARQL queries

1
2
3
14
1
15
80 ontologies were found that had these three
terms
By default, ontologies are ordered by their
popularity, but they can also be ordered by
recency or size.
Lets look at this one
16
Basic Metadata hasDateDiscovered  2005-01-17
hasDatePing  2006-03-21 hasPingState
 PingModified type  SemanticWebDocument
isEmbedded  false hasGrammar  RDFXML
hasParseState  ParseSuccess hasDateLastmodified
 2005-04-29 hasDateCache  2006-03-21
hasEncoding  ISO-8859-1 hasLength  18K
hasCntTriple  311.00 hasOntoRatio  0.98
hasCntSwt  94.00 hasCntSwtDef  72.00
hasCntInstance  8.00
17
(No Transcript)
18
rdfsrange was used 41 times to assert a value.
owlObjectProperty was instantiated 28 times
timeCal defined once and used 24 times (e.g.,
as range)
19
These are the namespaces this ontology uses.
Clicking on one shows all of the documents using
the namespace.
All of this is available in RDF form for the
agents among us.
20
Heres what the agent sees. Note the swoogle and
wob (web of belief) ontologies.
21
We can also search for terms (classes,
properties) like terms for person.
22
10K terms associated with person! Ordered by
use.
Lets look at foafPersons metadata
23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
87K documents used foafgender with a foafPerson
instance as the subject
27
3K documents used dccreator with a foafPerson
instance as the object
28
Swoogles archive saves every version of a SWD
its seen.
29
(No Transcript)
30
2

• An NSF ITR collaborative project with
• University of Maryland, Baltimore County
• University of Maryland, College Park
• U. Of California, Davis
• Rocky Mountain Biological Laboratory

31
An invasive species scenario

• Nile Tilapia fish have been found in a California
  lake.
• Can this invasive species thrive in this
  environment?
• If so, what will be the likelyconsequences for
  theecology?
• Sowe need to understandthe effects of
  introducingthis fish into the food webof a
  typical California lake

32
Food Webs

• A food web models the trophic (feeding)
  relationships between organisms in an ecology
• Food web simulators are used to explore the
  consequences of changes in the ecology, such as
  the introduction or removal of a species
• A locations food web is usually constructed from
  studies of the frequencies of the species found
  there and the known trophic relations among them.
• Goal automatically construct a food web for a
  new location using existing data and knowledge
• ELVIS Ecosystem Location Visualization and
  Information System

33
East River Valley Trophic Web
http//www.foodwebs.org/
34
Species List Constructor

• Click a county, get a species list

35
The problem

• We have data on what species are known to be in
  the location and can further restrict and fill in
  with other ecological models
• But we dont know which of these the Nile Tilapia
  eats of who might eat it.
• We can reason from taxonomic data (simlar
  species) and known natural history data (size,
  mass, habitat, etc.) to fill in the gaps.

36
(No Transcript)
37
Food Web Constructor

• Predict food web links using database and
  taxonomic reasoning.

In an new estuary, Nile Tilapia could compete
with ostracods (green) to eat algae. Predators
(red) and prey (blue) of ostracods may be affected
38
Evidence Provider

• Examine evidence for predicted links.

39
Status

• Goal is ELVIS (Ecosystem Location Visualization
  and Information System) as an integrated set of
  web services for constructing food webs for a
  given location.
• Background ontologies
• SpireEcoConcepts concepts and properties to
  represent food webs, and ELVIS related tasks,
  inputs and outputs
• ETHAN (Evolutionary Trees and Natural History)
  Concepts and properties for natural history
  information on species derived from data in the
  Animal diversity web and other taxonomic sources
• Under development
• Connect to visualization software
• Connect to triple shop to discover more data

40
UMBC Triple Shop
3

• http//sparql.cs.umbc.edu/
• Online SPARQL RDF query processing with several
  interesting features
• Automatically finds SWDs for give queries using
  Swoogle backend database
• Datasets, queries and results can be saved,
  tagged, annotated, shared, searched for, etc.
• RDF datasets as first class objects
• Can be stored on our server or downloaded
• Can be materialized in a database or(soon) as a
  Jena model

41
Web-scale semantic web data access
data access service
the Web
agent
Index RDF data
ask (person)
Search vocabulary
Search URIrefs in SW vocabulary
inform (foafPerson)
Compose query
ask (?x rdftype foafPerson)
Search URLs in SWD index
Populate RDF database
inform (doc URLs)
Fetch docs
Query local RDF database
42
Who knows Anupam Joshi? Show me their names,
email address and pictures
43
The UMBC ebiquity site publishes lots of RDF
data, including FOAF profiles
44
PREFIX foaf lthttp//xmlns.com/foaf/0.1/gt SELECT
DISTINCT ?p2name ?p2mbox ?p2pix FROM ??? WHERE
?p1 foafsurname "Joshi" . ?p1
foaffirstName Anupam" . ?p1
foafmbox ?p1mbox . ?p2
foafknows ?p3 . ?p3 foafmbox
?p1mbox . ?p2 foafname ?p2name
. ?p2 foafmbox ?p2mbox .
OPTIONAL ?p2 foafdepiction ?p2pix .
ORDER BY ?p2name
45
(No Transcript)
46
(No Transcript)
47
(No Transcript)
48
302 RDF documents were found that might have
useful data.
49
Well select them all and add them to the current
dataset.
50
Well run the query against this dataset to see
if the results are as expected.
51
The results can be produced in any of several
formats
52
(No Transcript)
53
Looks like a useful dataset. Lets save it and
also materialize it the TS triple store.
54
(No Transcript)
55
We can also annotate, save and share queries.
56
Work in Progress

• There are a host of performance issues
• We plan on supporting some special datasets,
  e.g.,
• FOAF data collected from Swoogle
• Definitions of RDF and OWL classes and properties
  from all ontologies that Swoogle has discovered
• Expanding constraints to select candidate SWDs to
  include arbitrary metadata and embedded queries
• FROM documents trusted by a member of the SPIRE
  project
• We will explore two models for making this useful
• As a downloadable application for client machines
• As an (open source?) downloadable service for
  servers supporting a community of users.

57
This talk

• Motivation
• Swoogle Semantic Websearch engine
• Use cases and applications
• Observations
• Conclusions

58
Will Swoogle Scale? How?

• Heres a rough estimate of the data in RDF
  documents on the semantic web based on Swoogles
  crawling

We think Swoogles centralized approach can be
made to work for the next few years if not longer.
59
How much reasoning should Swoogle do?

• SwoogleN (Nlt3) does limited reasoning
• Its expensive
• Its not clear how much should be done
• More reasoning would benefit many use cases
• e.g., type hierarchy
• Recognizing specialized metadata
• E.g., that ontology A some maps terms from B to C

60
A RDF Dictionary

• We hope to develop an RDF dictionary.
• Given an RDF term, returns a graph of its
  definiton
• Term ? definition from official ontology
• TermURL ? definition from SWD at URL
• Term ? union definition
• Optional argument recursively adds definitions of
  terms in definition excluding RDFS and OWL terms
• Optional arguments identifies more namespaces to
  exclude

61
This talk

• Motivation
• Swoogle Semantic Websearch engine
• Use cases and applications
• Observations
• Conclusions

62
Conclusion

• The web will contain the worlds knowledge in
  forms accessible to people and computers
• We need better ways to discover, index, search
  and reason over SW knowledge
• SW search engines address different tasks than
  html search engines
• So they require different techniques and APIs
• Swoogle like systems can help create consensus
  ontologies and foster best practices
• Swoogle is for Semantic Web 1.0
• Semantic Web 2.0 will make different demands

63
For more information
http//ebiquity.umbc.edu/
Annotatedin OWL