Textual-Geospatial Integration Project

1
Textual-Geospatial Integration Project

• JAMES FREW
• University of California, Santa Barbara

2
Geospatially-Augmented Search

• Whats here?
• Find library objects associated with a given
  location
• Place name(s)
• Footprint (geographic extent)
• Wheres this?
• Find the location(s) associated with a given
  library object

3
Examples (from TREC-9)

• Find documents that contain residential real
  estate listings within New Jersey.
• Find reports on automobile traffic in the
  Washington, DC metropolitan area.
• What forms of entertainment are available in
  Newport Beach, California?

4
Why Is GAS Difficult?

• Few library objects have explicit locations
• Assigned reliably
• Identified in objects metadata
• Many objects (especially text documents) have
  implicit locations
• Present in, or inferable from, objects content
• Not necessarily identified as locations

5
Wheres This Service
PARSE
text document
potential names, types, coordinates
type thesaurus
LOOKUP
gazetteer
gazetteer entries (known places)
ANALYZE
ranked footprints and placenames
best name(s)
EVALUATE
composite footprint
6
Geo-parsing

• Extract geographic facts from text
• Characterize by
• Potential place component
• name, type, footprint
• Related fact (with preposition)
• in , northeast of , etc.
• Frequency
• Importance
• Context

7
Geo-parsing Example (1/2)

• (California,,,,1,K)
• (Callahan,,,(in,California),1,K)
• (Callahan-Yreka,,,(area of,),1,T)
• (Early Cambrian,,,,1,B)
• (Klamath Mountains,,,(eastern,),1,T)
• (Klamath Mountains,,,(within,),1,B)
• (Klamath Mountains,,,,1,K)
• (Northern California,,,,1,T)
• (Ordovician,,,,1,B)
• (Ordovician,,,,1,K)
• (Paleozoic,,,(in,California),1,B)
• (Paleozoic,,,,1,K)

8
Geo-parsing Example (2/2)

• (Silurian,,,,1,K)
• (Siskiyou County,,,(in,California),1,K)
• (Skookum Gulch,,,,1,K)
• (Skookum Gulch,,,,1,T)
• (Skookum Gulch,,,,2,B)
• (United States,,,,1,K)
• (Yreka,,,(in,California),1,K)
• (,fault,,,2,B)
• (,rocks,,,6,B)
• (,,N410000N420000W1220000W1230000,,1,C)
• (,,,(in,North America),1,B)

9
Lookup Example Feature Type

• Fault partial match fault zones
• Rocks use natural rock formations

10
Lookup Example Gazetteer

• Place Name exact partial
• Skookum Gulch 1 0
• Klamath Mountains 1 0
• Northern California 1 0
• California 1 492
• Callahan 1 1
• Silurian 0 5
• Siskiyou County 1 14
• United States 1 273
• Yreka 1 12
• North America 0 8
• within footprint of California

11
Analysis Criteria

• Placement in document
• e.g. keywords, title gt body
• Frequency in document
• Exact match in gazetteer
• Accuracy of gazetteer footprint
• e.g. points lt bounding boxes
• Scale of gazetteer footprint
• Size of focus area / size of footprint

12
Analysis Example Results

• High confidence
• Callahan in California
• Yreka in California
• Skookum Gulch
• Klamath Mountains (eastern)
• Siskiyou County

• Low confidence
• Northern California
• United States
• North America

13
Evaluation Example
Skookum Gluch
Klamath Mountains
California
Callahan in California
Siskiyou County in California
United States
Yreka in California

• Additional placenames
• Shasta Butte City
• Yreka City
• Thompson's Dry Diggings

14
Whats Here Service
Query Expansion
Gazetteer
Example Query Bodies of Water near
Chicago Expansion Terms Lake Michigan, Chicago
River
15
Manual Relevance Feedback
Spatial Synonyms Chicago, IL Chicago River
AIRE
UserInterface
Gazetteer
Query
Place Names Chicago
Query Parser
16
Automatic Relevance Feedback
AIRE
Spatial Query Results Chicago River, Lake
Michigan
Gazetteer
RF System
Place Names, Surrounding Type Terms Bodies of
Water
Expanded Query
Document Ranker
17
Whats Here Components

• Place names ? footprints
• Requires place name ranking scheme
• Chicago, IL gt Chicago tectonic plate in Brazil
• Type terms ? classes
• Requires class thesaurus API
• Bodies of Water ? Water Bodies
• 3. Gazetteer ? spatial synonyms
• Requires gazetteer API results ranking
• Bodies of Water near Chicago ? set of gazetteer
  queries

18
The Light at the End of the Tunnel

• You submit
• a document
• You get
• a place
• Best
• Also-rans
• Alternatives
• What you do with this is your business

19
Brought To You By

• UCSB
• Linda Hill
• Greg Janée
• Dave Valentine
• Satoshi Ikeda
• (Japan Patent Office)

• IIT
• Steven Beitzel
• Ophir Frieder
• David Grossman
• Eric Jensen
• Vasif Shaikh

20
(No Transcript)