BUILDING NANOBANK

1
BUILDING NANOBANK

• Data Structure and Selection Criteria
• Jason Fong and Emre Uyar
• University of California, Los Angeles

2
What is Nanobank?

• Nanobank is a collection of observations from
  various sources (scientific articles, patents and
  government grants), determined to be related to
  nanotechnology field, either by probabilistic
  information retrieval (IR) methods or by being
  declared nano by a source authority.

3
Data Sources - Articles

• 580,711 scientific articles from peer reviewed
  journals.
• Source Science Citation Index, Arts Humanities
  Citation Index and Social Sciences Citation Index
  of the Institute for Scientific Information Inc.
  (ISI). All together, these indexes contain more
  than 24,250,000 entries from over 8,700 peer
  reviewed scientific journals.

4
Data Sources Patents and Grants

• 240,437 patents from U.S. Patenting and Trademark
  Offices online database of more than 4,000,000
  patents, granted by USPTO from 1976 to 2006.
• 52,831 grants from NIH and NSF databases.

5
Data Contents

• Articles
• Titles
• Journal volume and issue numbers
• Publication years
• Author names
• Names and addresses of organizations affiliated
  with authors

6
Data Contents

• Patents
• Titles and abstracts
• Application and grant dates
• Names and addresses of inventors and assignees
• U.S. and international patent classifications

7
Data Contents

• Grants
• Titles and abstracts
• Receiving organization names and addresses
• PI and co-PI names
• Grant amounts

8
Nanobank Data Structure

• Internal database
• Stored in a relational database
• Separate tables for various data items
• ID numbers for each item link between tables
• Version posted on Nanobank.org
• Denormalized form of internal database
• Storing redundant data isnt as space-efficient,
  but lessens the need to join multiple tables
• Nanobank Codebook contains detailed information
  on tables and fields available in each

9
Document Selection

• Document Selection Methods
• Keywords
• Probabilistic
• Authority-selected
• Tables include a field to indicate selection
  method
• nanobank_flag 1 if selected by Keywords or
  Probabilistic 0 otherwise
• authority_flag 1 if Authority-selected 0
  otherwise

10
Document Selection Keywords

• Search for text patterns matching words or
  phrases related to nanotechnology
• Words and phrases chosen by subject specialists
• Less effective for identifying very early or
  recent documents
• Early documents were written before the terms
  were in common usage
• Recent documents have terms that are too new to
  be included in the search patterns

11
Document Selection Probabilistic

• Incorporates new terms as they come into common
  usage
• Uses the Xapian search engine library to perform
  ranking calculations
• Analyzes document text and ranks against a set of
  query terms

12
Document Selection Probabilistic

• Initial query terms from the Virtual Journal of
  Nanoscale Science Technology (VJNano)
• All articles in VJNano assumed to be relevant
• Select highest ranked terms
• Document selection process
• Use initial query terms to select relevant
  documents from all journal articles
• Select additional terms from those relevant
  documents and add to query
• Repeat selection with expanded query terms

13
Document Selection Authority Set

• Articles
• Listed in the Virtual Journal of Nanoscale
  Science Technology
• Patents
• Listed under United States Patent Classification
  Class 977 (Nanotechnology)
• NSF Grants
• program name contains nano
• NIH Grants
• NIH descriptive tag contains nano

14
GEOCODING

• Standardizing between differing naming
  conventions used in different sources.
• Standardizing between non-uniformity in how
  observations are recorded.
• Correcting common mistakes.
• For US observations Providing different grouping
  units (other than city and state) not available
  in original data sources, like counties and BEA
  areas.

15
COUNTRY GEOCODING

• Country names in all observations are cleaned,
  standardized and assigned an ISO code (2 digit
  alphabetical)
• Current ISO list of countries is taken as basis
  historical entries assigned to the closest
  current country to the extend available.

16
US GEOCODING

• US observations are those in 50 US states, DC and
  7 US associated areas.
• Cities, states, counties and BEA economic areas
  are coded using Populated Places data obtained
  from FIPS 55 database and BEA.
• Basis is the city-state combination. City names
  are standardized and matched to the names in FIPS
  database on a state-by-state basis.
• In articles, 99.98 of US observations have been
  assigned a definite city - state code.

17
US GEOCODING Variables Created

• Standard_city_name Standardized name as it
  appears on the FIPS database (corrected for
  misspelings, abbreviations, etc...)
• State_code 2 digit numeric code.
• City_code 5 digit numeric code, unique by state.
• County_code 5 digit numeric code.
• County_name
• City code state code uniquely determine a
  populated place.
• Numeric codes are same as the codes used by FIPS.

18
GEOCODING US BEA Areas

• Bureau of Economic Analysis (BEA) created 179
  Economic Areas in the US by asigning each county
  is assigned to a unique BEA.
• BEA_code 3 digit numeric code that determines
  the associated BEA Economic area for each
  observation.
• "BEA's economic areas define the relevant
  regional markets surrounding metropolitan or
  micropolitan statistical areas. They consist of
  one or more economic nodes - metropolitan or
  micropolitan statistical areas that serve as
  regional centers of economic activity and the
  surrounding counties that are economically
  related to the nodes.
• The economic areas were redefined on November 17,
  2004, and are based on commuting data from the
  2000 decennial population census, on redefined
  statistical areas from OMB (February 2004), and
  on newspaper circulation data from the Audit
  Bureau of Circulations for 2001."

19
ORGANIZATION CODES

• Each observation is assigned an alpha numerical
  code.
• 2 digit alphabetical part determines the
  organization type.
• Numeric part groups names that are same up to
  standardization and hand cleaning

20
Organization Codes Types of Cleaning

• Standardization of common identifiers
• IBM IBM Corp. IBM Corporation
• Univ University University of Universidade
  Universidad Univerzitet Universita
  Universitat Universiti Universite
  Universitet Universiteit
• Using look up tables and hand cleaning to
  identify common variants (and misspellings) of
  names used by the same organization
• IBM Int Buisness Machines
• International Business Machines Corporation
• Int Business Machines Operation