Mapping Knowledge Domains: A Localized Approach

1
Mapping Knowledge Domains A Localized Approach

• Xia Lin
• Howard D. White
• Jan Buzydlowski
• Chaomei Chen
• Drexel University
• Philadelphia, PA

2
Abstract

• A prototype system, PNASLink, has been developed
  to investigate the potential of localized
  knowledge-domain mapping. The localized approach
  centers on individual concepts, authors,
  documents, or journal titles, any one of which
  can be used as a seed term to initiate mapping.
  The mapping displays the seed term and its
  associated terms in semantic neighborhoods to
  reveal the domains intellectual structure.
  PNASLink generates these maps on the fly. The
  user can click on any item in them to generate a
  new map. These related neighborhood maps help
  the user understand the knowledge domain. They
  also serve as guides to assist in query
  construction and online searching. The PNASLink
  Web interface permits retrieval of records and
  full texts from bibliographic databases on the
  basis of mapped terms.

3
PNASLink Architecture
Noah indexing engine
Mapping Algorithms
Java Application Server
PNAS Online
Map Displays
4
PNASLink Features

• Generating Maps on the fly
• For a given
• Term
• Author
• Journal
• From PNAS data sets
• of more than 1,500,000 citations
• and more than 600,000 terms

5
Mapping Processing

• For a given term
• Find 24 terms co-occurred most often with the
  given term
• Drive a 25 by 25 data matrix
• Apply the two mapping algorithms to the data
  matrix
• Generate the graphical display
• All are done instantly.

6
PNAS Features

• User-centered Interaction
• Users can select from the display a new
  focused term for the next mapping
• Users can remove some terms from the display and
  generate the map display again
• Users can apply a filter term to re-generate
  the map displays

7
PNASLink Implementation

• Mapping
• Two Algorithms
• SOM -- Kohonen Self-Organizing Feature Maps
• PFNET -- Path Finder Networks
• Multiple views
• Author/Terms/Journals
• Interaction
• Providing means for the user to go from one
  neighborhood to another.
• Connecting visual displays to search engines.

8
(No Transcript)
9
Author Mapping

• Based on Co-citation Analysis
• The more often two authors are co-cited, the more
  likely their work are in the same research areas.
• Pair-wise analysis of a group of authors who are
  often cited together reveals the most salient
  intellectual structures of the areas for which
  these authors represent.

10
(No Transcript)
11
Map of Michael Bishop(After Tom Maniatis was
removed)
12
Concept Mapping

• Based on Term Co-occurrence Analysis
• Patterns how terms are co-occurred in documents
  will reveal semantic relationships of the terms.
• The co-occurrence patterns of a group of terms
  will reveal semantic structures of a topic, a
  field, or a domain.

13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
Journal Mapping

• Based on journal co-citation patterns, journals
  are clustered to reveal their similarities in
  content and coverage.

19
(No Transcript)
20

• After Science, Nature, Cell, and J-BIOL-CHEM are
  removed.

21
(No Transcript)
22
Mapping Knowledge
23
Domains
24
URLhttp//project.cis.drexel.edu/pnas/
25
A Localized Approach