Alexandria

1
Alexandria Digital Library Project
University of California, Santa Barbara
http//www.alexandria.ucsb.edu/
2
Textual- Geospatial Integration Project
Aerial photos
Maps
NSF National Science Digital Library
Project 2001-2003
3
Project Goals

• Extend NSDL infrastructure by enabling
• geographic queries
• for text and non-text items across heterogeneous
  digital libraries
• geographic referencing
• of arbitrary texts without explicit geographic
  cataloging

4
Participants

• University of California,Santa Barbara
• James Frew, PI
• Terence Smith
• Michael Bueno
• Linda Hill
• Information Retrieval Lab, Illinois Institute of
  Technology
• Ophir Frieder
• David Grossman
• Eric Jensen

The American Geological Institute (AGI) has
permitted us to use a set of their GeoRef records
for system training.
5
Geospatially-

• 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

6
Augmented Search

• Examples Queries 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?

7
The stages
lookup in gazetteer

• gazetteer entries
• names
• footprints

Oral histories

• georeferenced facts
• placenames
• IN
• ENVIRONS
• PIECE OF
• feature types

spatial analysis
identify best footprint
geo-parsing
8
The evaluation
9
Example Text

• title Stress-induced borehole elongation a
  comparison between the four-arm dipmeter and the
  borehole televiewer in the Auburn geothermal well
  
• keys applications Auburn borehole breakouts
  boreholes caliper logging Cayuga County New
  York deformation dipmeter logging
  elongation field studies fractures
  geophysical surveys instruments New York
  patterns preferred orientation rock mechanics
  spallations stress structural analysis
  surveys televiewers United States
  well-logging
• abstract The nature and origin of borehole
  elongation recorded by the four-arm dipmeter
  calipers is studied utilizing information
  obtained from hydraulic fracturing stress
  measurements and borehole televiewer data taken
  in a well located in Auburn, New York. A
  preferred orientation N10 degrees W-S10 degrees
  E, or -10 degrees and a less prominant E-W
  orientation of borehole elongation, was observed
  on two runs of the dipmeter. Comparisons of
  borehole geometry determined using the televiewer
  and the dipmeter show that both tools give the
  same orientation of borehole elongation provided
  that the zone of elongation is longer than 30 cm.
  Comparisons of dipmeter caliper data with
  orientation of in situ stress and natural
  fractures, obtained from hydrofracturing tests
  and televiewer data show that the N10 degrees
  W-S10 degrees E borehole elongations (1) are
  axisymmetric, (2) are aligned with the minimum
  horizontal stress S (sub h) and (3) are not
  associated with natural fractures intersecting
  the well. These elongations are interpreted as
  stress-induced well bore breakouts. The E-W
  elongation direction is characterized by an
  assymmetric borehole cross section in thinly
  bedded rocks and is not caused by breakouts. This
  assymmetric geometry can be discriminated from
  breakouts using the oriented electric
  measurements provided by the dipmeter. This study
  demonstrates that the dipmeter can be used to
  determine the orientation of S (sub h) confirming
  the results of earlier less detailed studies, and
  provides a firm basis for mapping regional stress
  patterns using existing dipmeter data.--Modified
  journal abstract
• GeoRef bibliographic record from the TGI test set
  of 7523 records

10
Manual Analysis
11
Geoparsing
fact (name?, type?, footprint?,
related-fact?, certainty, importance)
blue valid fact green partially valid
fact red invalid fact

• Geoparsing scoring
• valid fact 1
• partially valid fact 0.25
• invalid fact 0

• Geoparsing performance
• parser recall 4.25/6 0.71
• parser precision 4.25/8 0.53

12
Gazlookup

• operator equals (exact match)
• auburn .. 37 entries
• new york .. 18 entries
• united states .. 1 entry
• cayuga county .. 1 entry
• auburn new .. 0
• county new .. 0
• york .. 50 entries
• TOTAL 105

• Gazlookup performance
• lookup recall 3/4 0.75
• lookup precision 3/105 0.03

13
Scatter of points
Scatter of 105 points from equals Gazlookup
Baseline clustering
Clustered points (67) in the US and Canada
14
Derived footprint
GeoRef footprint
Derived footprint from points
Very low spatial similarity between TGI box and
reference box from GeoRef
Footprint for equals lookup data and simple
clustering, compared to GeoRef footprint
15
Statistics redux

• Based on comparison of automated processes to
  manual analysis and GeoRef box for one sample
  record
• Geoparsing
• Recall .. 0.71
• Precision . 0.53
• Gazlookup
• Recall .. 0.75
• Precision . 0.03
• TBI bounding box
• Recall .. 0.75
• Precision ... 0.05
• Similarity to reference .. 0

16
Next steps

• Set new conditions
• Find settings that give good results for 10 test
  records
• Run 7,524 GeoRef test records through TGI
• Calculate similarity of TGI boxes to GeoRef boxes
• Choose 10 new test records for manual analysis
  from best worst results
• Reset conditions
• Repeat