InfoVis 2004 Contest
InfoVisExplorer

Contest webpage:

Authors and Affiliations:

• Jaroslav Tyman, Georgia Institute of Technology, dabel@cc.gatech.edu
• Grant P. Gruetzmacher, Georgia Institute of Technology, gte133q@mail.gatech.edu
• John Stasko, Georgia Institute of Technology, stasko@cc.gatech.edu

Tool(s):

Three tools, developed by the authors, were used to translate the XML data, group keywords into topics, and visualize the results. We named the tools XMLImport, InfoVisBrowser, and InfoVisExplorer. These were developed for OS X using Cocoa and OpenGL. We discuss only the visualization tool here, InfoVisExplorer.

TASK 1: Static Overview of 10 years of Infovis

• Process:
  We decided to visualize the articles in three dimensions. The axes represent years, number of interesting elements, and grouping in topics. The number of interesting elements is simply the addition of the number of authors, keywords, and references in the article. This is the only axis on a logarithmic scale. The grouping of topics was done by assigning each topic a position on the graph. Keywords were grouped into 5 topics: data structures, HCI, implementation, techniques, other. Each paper could span several topics. The topic values of all the keywords in a paper were averaged and the paper placed on that location.
  Authors were placed on a plane above the article plot. An author's placement is the average projection of their papers to the author plane. Jitter was introduced to both articles and authors since there were several cases of multiple overlap.
  Finally, colors were assigned to topics, and bubbles drawn around each element.

• Image 1.1 :
  
• Insight:
  This image shows the proportion of topics across the past 10 years.
• Caption for exhibit:
  The colors in the above image reflect the topics in InfoVis.

TASK 2: Characterize the research areas and their evolution

• Process:
  This task was the main reason we chose to group keywords into topics. Once we had our keywords organized, we created an interface element that can turn on and off the display of certain topics. To enhance the visuals, we draw translucent spheres around each paper. These radius of these spheres can be adjusted quickly and smoothly, and the resulting views create vastly different visual effects. To facilitate quick responses, the program precomputes as much information as possible leaving only the OpenGL calculations to perform.
  To explore the research areas interactively, we enable the possibility of drawing links between papers matching the filters. The user can identify key papers and their links throughout the chart.
• Image 2.1:
  
• Insight 2.1:
  Here we can see the gradual introduction and growing popularity of certain topics.
• Image 2.2:
  
• Insight 2.2:

  Here, we have only data, implementation, and techniques highlighted. We can see that first keywords related to implemenation details were used, then data structures, and finally a discussion of techniques became the proponderance of keywords.

TASK 3: The people in InfoVis

Task 3.1: Where does a particular author/researcher fit within the research areas defined in task 2?

• Process:
  The methodology here is similar to the ones already described, however, now we use the additional view of authors along the top plane of the chart. We accomplish this task with the links between articles, links between authors, and color coded topic spheres.
  There are several ways to inspect an author such as G. Robertson. First, we select his name in the list of authors along the right, "George G. Robertson." Then, we: look at paper links, look at colors, adjust bubbles, show author links, show connected author links.
• Image 3.1.1:
  
• Insight:
  Here we've highlighted George G. Robertson. We can see the papers of his that included keywords, and we can see at the top of the image his place on the author projection (most of his keywords dealt with data, so his bubble is blue).

• Image 3.1.2:
  
• Insight:
  Here we have shown references to articles written by Mr. Robertson. We can clearly see his papers have been referenced many times. We can also see some heavily referenced papers are in the bottom area of the visualization. They are located here because they lacked keywords in the dataset.

Task 3.2: What, if any, are the relationships between two or more or all researchers?

• Process:
  The process here is very similar to the one above. In this case however, we visualize links between authors. The first link type is between coauthors, calculated during import and stored in a lookup table. After the co-author table is built, we construct fully connected sets of authors. By using integer indexing, the traversals are fast, and the connected-set table takes advantage of the data in the co-author table to improve performance.
  Once again we use OpenGL to draw the scene. With all of the positions, colors, and other attributes calculated the visualization remains fast and smooth.
• Image 3.2.1:
  
• Insight:
  This is a closeup of the author projection on the top of the display. Links are drawn to Mr. Robertson's co-authors.
  
• Image 3.2.2:
  
• Insight:
  
Here we show the complete connected set of authors for Mr. Robertson. We can see that Mr. Robertson's group is very well connected and spans across many topics.


• Image 3.2.2:
  
• Insight:
  For comparison, we show the complete connected set of authors for John Stasko. We can see that his group is somewhat smaller, and covers the topic of techniques more heavily than Mr. Robertson's set.

TASK 4: Find Errors in the Data Set

• Process:
  During the early stages of the project, we realized that the dataset contained many errors, either omissions or erroneous data. We decided to facilitate finding some of the errors through the use of our tool. The process here is similar to those above. We plot articles with no keywords below the chart, and we can selectively filter which references to display.
  
• Image 4.1:
  
• Insight 4.1:
  The articles below the chart lack any keywords at all, an omission in the data set.
  
• Image 4.2:
  
• Insight 4.2:
  All articles from 1974 to 1996 are in the filtered set, references from articles are shown only to articles outside the filter. What we see here are several references going into the future. Upon close inspection, we can see that Mr. Robertson's own paper, The Document Lens, is one of the errors in the data set. It is listed as being published in 1999, yet most papers reference it as being published in 1993.

Web Accessibility