Disciplinebased Visualization

1
Discipline-based Visualization

• Howard D. White
• Xia Lin
• Katherine W. McCain
• Drexel University

2
Comparing 2 Views of Information Science Authors

• Produced with similar (not identical) data sets
  from same research project.
• Lins is a Kohonen feature map.
• White and McCains map results from SPSS
  multidimensional scaling and clustering.

3
Lins Kohonen map

• Represents full set of 120 most-cited authors in
  information science, 1972-95, based on how they
  are co-cited with each other.
• Shaded areas show subject regions. Neighboring
  regions are intellectually close.
• Input was Pearson r correlation coefficients for
  each author pair--the measure of similarity
  between authors.

4
White McCains MDS Map

• Shows 100 most cited authors in information
  science for 1988-95, based on how they are
  co-cited with each other.
• Different from Lins map because of SPSS
  limitations not 120 authors not 1972-95.
• Is a re-oriented version of map in White McCain
  article in April 1998 JASIS.

5
White McCains MDS Map (cont.)

• Enhanced with lines that reflect a
  complete-linkage clustering of authors.
  Boundaries show broad subject regions.
• SPSS programs used were ALSCAL for
  multidimensional scaling and CLUSTER for
  complete-linkage clustering.
• Input was Pearson r correlation coefficients for
  each author pair--the measure of similarity
  between authors.

6
RESULTS

• Kohonen algorithm and multidimensional scaling
  produce broadly similar maps, both highly
  intelligible.

7
Both Maps

• Isolate two main sub-disciplines of information
  science INFORMATION RETRIEVAL (top
  half) DOMAIN ANALYSIS (bottom half)

8
Clockwise in both maps

• Top left Hard retrievalists-- quantitative
  retrieval theory, algorithmic solutions to
  document retrieval problems, IR systems
  evaluation
• Top right Soft retrievalists-- practical
  aspects of online searching, library
  automation, user cognition and behavior

9
Clockwise in both maps (cont.)

• Midright Theoretical gurus --external to the
  discipline
• Lower right/bottom Domain analysts
  --specialists in scientific and technical
  communication, sociology of science, studies
  of specific literatures, often through citation
  analysis

10
Clockwise in both maps (cont.)

• Lower left Bibliometricians-- mathematical
  modelers of properties of literatures
• Midleft Dual contributors-- feet in both
  camps, bibliometrics and information retrieval

11
Both Maps

• Reveal clusters of authors in specialties around
  the edges and relatively few authors in central
  region. (Latter implies that not many authors
  are seen as integrating IS specialties.)
• Show much the same progression of specialties
  around the edges.

12
Both Maps

• Have the largely the same, good transitional
  authors between specialties.
• Show retrievable literatures represented by a
  display of interrelated authors.

13
BUT

• Kohonen feature map has advantages over
  cluster-enhanced MDS map.
• It took 6 minutes to create on a Sun workstation,
  given the input correlation matrix.
• It can accept potentially far more authors than
  ALSCALs limit of 100 or the 120 in Lins present
  map.

14
Further advantages

• It is an electronic document that can rapidly be
  developed, through Java programming, as an
  interface for retrieval.
• In Lins present Website version, visitors can
  draw a rectangle with the cursor around groups of
  author-points. This produces a pop-up window
  with the authors full names. When these are
  clicked on, an Alta Vista search is launched for
  documents related to the author on the Web (e.g.,
  his or her home page).

15
Implications

• Future maps could be developed to retrieve
  valuable caches of documents by or about authors,
  or documents citing them.
• Might become part of electronic publishing
  would assist in retrieving documents from the Web
  or on CD-ROMs.

16
The Java Interface

• Interactive
• control the number of authors shown on the screen
  
• control the number of Labels on the screen
• Labels are currently defined manually
• select authors with mouse drags
• can also type in names to search

17
The Java Interface

• Distributed
• It is an applet on the web
• It is a front-end to web-search engines
• Clicking on an authors name will retrieve the
  authors related work available on the web
• It can be font-ends to multiple databases

18
Interface Design Issues -1

• If an author is involved in three subject areas,
  where should he/she be located?
• A. In one major area selected
• B. In the middle of the three areas
• C. In all the three areas
• The algorithms typically choose answer B.
• Users like answer C.

19
Interface Design Issues -2

• When there are many authors in an area
• A. Display all the names
• B. Display selective names only
• C. Display dot icons only
• Names would overlay with each others in Answer A.
• Answers B and C both lose some visual power.

20
Interface Design Issues -3

• How to draw boundaries of subject areas?
• A. Draw clear lines between subject areas.
• B. Leave some blurry areas among subject areas.
• C. Never clearly specify the subject areas.
• Subject areas are in the eyes of gazers.
• Depends on the data and the users.

21
Interface Design Issues -4

• How details should the labels be?
• Should the map serve as
• A. the table of contents
• B. the back-of-the-book index
• C. both
• Users seem to want both.
• But a map cannot be everything.

22
Future Research

• To continue exploring algorithms for
  literature/discipline mapping
• automatically identify keyword labels/subject
  areas
• combine data of keyword indexing and citation
  indexing
• use multi-discipline data
• reduce mapping time
• from 6 minutes to 6 seconds?

23
Future Research

• To continue developing the interface
• Better graphical layouts and visual design
• more interactive functions
• multi-levels of displays
• user studies

24
Future Research

• To improve the connection between the map
  interface and databases
• Increasing search precision
• Create our own databases for ISI data
• Add other authors to the map
• Every one of you should be on the map
• if you cite any of the authors on the map, and
• if a live calculation of your citation patterns
  can be done.

25
URL for the map Interface

• http//
• research.cis.drexel.edu/
• citation/index.html

26
Describing a Discipline

• By subject areas/topics
• By people involved
• By keywords
• By its relationships with other disciplines
• By dimensionality or structures of the discipline
• By all of the above in a map?

27
(Online retrieval)
Retrieval
User
(IR theories)
(Communication)
(General)
Citation
Document
(Bibliometrics)
28
Results of Kohonen Mapping of the Data

• Overall structures are similar to MDS map
• Labels generated through factor analysis can be
  easily put on the map
• Clusters identified on MDS map are also visible
  on Kohonen map
• Many more identifiable sub-areas
• Color codes for sub-areas
• More than two dimensions identified

29
Where are they located?

• Donald Kraft
• Edward Fox
• Raya Fidel
• Eugene Garfield
• Robert Hayes

30
Where are you located?

• Every one of you should be on the map
• if you cite any of them
• if we can do a live calculation of your
  citation patterns
• currently under development

31
Example
Online Searching
Searching Behaviors
Information Retrieval
Question Where should Nick Belkin be?
32
Answer
He should be in all the three, and more!
Online Searching
Borgman
Belkin
Fidel
Meadow
Searcher Behaviors
Information Retrieval
Ingwersen
...
User Theories
IR Theories
33
Future development
34
The First Set of Data

• 120 highly cited authors in Information Science
• A matrix of 120 by 120 of their co-citation
  counts

35
Results of Multidimensional Scaling on the data

• The map shows a clearly intelligible framework
  for IS
• Two sub-disciplines
• Clusters within the disciplines
• Two dimensions

36
Users
Documents
soft domain specialists
hard domain specialists