InfoVis 2004 Contest
MONKELLIPSE

Contest webpage: http://steel.lcc.gatech.edu/~dmccolgin/classes/infovis/Project/MonkEllipseV1.html

Authors and Affiliations:

• Lee Inman Farabaugh , Georgia Institute of Technology, lee@cc.gatech.edu
• Tzu-Wei Hsu, Georgia Institute of Technology, thsu@cc.gatech.edu
• Dave McColgin, Georgia Institute of Technology, davemail@cc.gatech.edu
• Kevin Stamper, Georgia Institute of Technology, kstamper@cc.gatech.edu

Tool(s):

We used PERL extensively, both to find additional content for the database, and to develop our categories and classify authors and papers. Our subsequent database is in PostgreSQL. The final program can run in any browser with the Flash plugin, and was created with Flash MX2004 and ActionScript 2.0. Flash communicates with the PostgreSQL database using an intermediate PHP page.

TASK 1: Static Overview of 10 years of Infovis

• Process :
  Research areas, or topics, were derived by first parsing out all of the keywords and counting the number of times they occurred. Next, we printed all of them out, cut them up, and manually arranged them into semantic groups. We defined each group with a tentative topic name. Topic names came from past topics in the InfoVis conference listings, lecture titles from Dr. John Stasko's class in InfoVis, and from groups of keywords that suggested a collective heading. We iterated this process until we arrived with a final set. Our database contains each portion of a document: date, abstract, references, etc. It also holds our derived categories and a list of all authors. MonkEllipse arranges each document title dynamically in an ellipse. Clockwise from the top, the documents are listed in chronological order. Shaded sections with year indicators provide further information regarding chronology. The derived topics are scattered in the center and sized according to the number of documents within them. Documents and topics affect one another when selected if they are related. In developing support for interactive tasks, we wanted to minimize both the number of 'modes' and the changes to the function or appearance of visual elements. It is easier for the interactor to learn and predict the effects of his actions with a single flexible display than it is to learn a number of task-specific modes and interactions. Early on, we noticed that this approach produced a pleasing display that could statically show chronology, evolution, and the names and relative influence of our derived research areas.
• Image 1.1:
  
• Insight :
  This image is of the system before any interaction occurs. The documents are arranged chronologically around the ellipse, clockwise from the top. Research areas, derived from InfoVis Conference listings of the past and Dr. John Stasko's InfoVis lectures, populate the center of the ellipse. Documents were categorized into the derived topics based on keywords or abstracts, dependent upon whether or not keywords were available. The size of the topics displayed in the center of MonkEllipse conveys the relative size and influence of each research area within Information Visualization. The year indicators surrounding the document titles show the distribution of publications within each year.
• Caption for exhibit:
  InfoVis documents from the past 10 years are arranged chronologically in an ellipse. Research areas are displayed at the center of the documents they abstract. Their size shows the relative number of documents within each research area, and thus, the roots and direction of the field as a whole. Year indicators show the distribution of publications for each year.
• Image 1.2:
  
• Insight :
  As discussed above, screenshots of actual interaction can provide visually interesting and informative still images. This allows the interaction itself to come through in the image, adding another layer to the evolution of Information Visualization, which the system explores. In the image above, the interactor searched for a specific author and document, then clicked on the returned document (in red). The green documents are those referenced by the red document; the green topics are those it falls under. The bottom pane displays document-specific information. Underlying all of this, every single InfoVis document of the past 10 years is arranged chronologically in an ellipse: an elegant, textured representation of the field. The background shading of years indicates years where research in InfoVis was more or less popular. Our research areas are displayed at the center of the documents from which they are derived . Their size shows the relative number of documents within each research area, and thus, the roots and direction of the field as a whole.
• Caption for exhibit :
  InfoVis documents from the past 10 years are arranged chronologically in an ellipse. Our research areas are displayed at the center of the documents from which they are derived. Their size shows the relative number of documents within each research area, and thus, the roots and direction of the field as a whole. Year indicators show the distribution of publications for each year.

TASK 2: Characterize the research areas and their evolution

• Process :
  In order to characterize the research areas, we came up with a set of topics under which we thought each of the articles could be classified. Our basic method is described above. We arranged these topics in the center of our visualization. Their size shows the relative number of documents within each research area, and thus, the roots and direction of the field as a whole. In order to see the articles that fall under a given topic, the interactor clicks on its name, which highlights all the articles with related keywords (the abstract was used if keywords were missing). The article titles are positioned according to an ellipse formula in chronological order around the topics. Thus, article highlighting allows the user to see temporal trends in the research area, such as the first paper published, clusters of more prolific publication in specific years, or a gradual increase in publications over time. Of course, research areas can also be characterized by their contributors. When a topic is selected, researchers who are published in that topic are shown. Their order is arranged according to our calculated Influence Factor (IF), a number based on the authors' publications in the field and the number of times they are referenced. When a document is selected, all the topics it falls under are highlighted. With the current infrastructure, one could also add support for selecting multiple topics, highlighting documents with different effects to show their combined influence as well as the amount of overlap between topics. Topics would then be toggle switches--once to select, again to deselect. We have prototyped several ways of supporting multiple topic selection, but have not had the time to implement them.
  The most pervasive problems we encountered were inconsistencies and omissions in the dataset. Some documents had little more than a title, and we did our best to gather further information. Many authors and keywords were spelled differently due to inconsistent punctuation or errors. Documents with such problems were more difficult to accurately classify with our parser algorithm. We tried many different color schemes to ensure that all of the screen objects are visisble and that hilighted objects stand out. The dark background allowed different colors to be more easily distinguishable from one another and still have unselected objects clearly visible. Another problem we encountered dealt with maximizing screen real estate. With 614 article titles arranged on a single screen, we had to experiment with size and rollover effects to ensure individual items could be identified and selected as our current version shows. Some article titles continue off-screen when rolled over, and so we added an echo of the rolled over title just beneath the ellipse, even matching size and color. Finally, because we wanted to keep the same basic interface for a variety of tasks, we had to find space to display the authors within a selected topic, the articles in that category, and keep all the topics available for interaction. We accomplished this with an animated change in the topic layout based on the user's actions. When a topic is selected, all topics moved quickly to a position that leaves room for the list of authors in the topic. The animation begins very quickly and slows down at the end so that the interactor perceives a fluid, understandable change. Unfortunately, the animation only displays as we intended if, once an action to change the layout is taken, the cursor is moved to the very edge of the screen. Otherwise, we believe mouse events fire behind the scenes that slow the animation considerably. When the author list is not needed, the space is reclaimed in order to leave only appropriate interactive objects onscreen, as well as improving aesthetics.
• Image 2.1 :
  
• Insight 2.1 :
  The bottom pane that holds title, authors, date, abstract, and a direct web link to the document in the ACM library, was populated due to a previous document selection. To support exploration with this information still present, this data is not removed or replaced until another document is selected. The interactor has clicked on the topic "Pan/Zoom." As a result, the database returned all related articles and all authors who have published in this area. The related articles have turned green and are slightly larger to assist accurate rollover and selection. Having the entire 10 year history in one view takes advantage of the our ability to see clusters and trends in data. In this image, there is a noticeable increase in publications over time, starting with the first publications in 1994 and 1995 (on the right), which appear to be the birth of this topic. Rolling over the first 3 articles highlighted in the 1994 section shows that "Galaxy of News...", "Powers of Ten Thousand..." and "Pad++" may be preliminary work in the area. Visually following the ellipse clockwise, it is clear there are progressively more publications in that area as time progresses. The left side of the ellipse reveals an increased number of green articles that are consistently spread out with a few articles each year from 1997 to present, indicating the establishment of Pan/Zoom as a valid research topic within InfoVis since its origination. Shneiderman and Bederson are leaders in the field, as evident from their top position in the list and their Influence Factors. Indeed, clicking through the articles in the topic shows that Bederson authored one of the first three articles.

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 task of discovering where a particular author/researching fits within the research areas is difficult, as there are a multitude of researchers in InfoVis and many of them are involved in multiple research areas within InfoVis. As our system stands now, the simplest way to find a particular author/researcher is to search the author's name. The results of this search return all articles associated with the researcher, regardless of whether the researcher was an author of the article, referenced by the article, or mentioned in the abstract of the article. Looking just at this view, while we can not garner where the researcher fits within the research areas, we can get an idea of how long this particular researcher has been involved in the field of Information Visualization, which is interesting in and of itself. Continuing on, clicking on any of the returned articles reveals the research area(s) into which the article fits. Based on the relationship between the researcher and the article, we can then determine the relationship between the researcher and the returned research areas. Sometimes, we are not interested in which research area a particular researcher fits, but rather, which researchers are involved in a particular research area. The process to accomplish this task is simple: a quick click on the topic of interest will return a list of authors who have contributed to the chosen research area.
  One final note: In considering different ways to accomplish this task, we entertained the idea of having an author view in MonkEllipse. The author view, put simply, would be almost an exact opposite to the topic view currently implemented. With the authors in the center of the ellipse, you would then be able to click on an author to return all articles by that author, along with a list of research areas to which the author as contributed. However, in prototyping the author view, we found a great increase in complexity, mostly due to the massive number of authors/researchers in the field of Information Visualization. Thus, this view has not been implemented; however, we believe that with the intricacies worked out, the author view of our system would serve this task well.
• Image 3.1.1 :
  
• Insight :
  Results returned from searching Robertson reveals that George Robertson has been a consistent contributor to the field of InfoVis, in research and in influence, almost since its origination (1989). Scanning through the returned results, one finds that in addition to a continued contribution to the field of InfoVis in general, Robertson's research is not limited to a specific topic in Information Visualization, but adds to the field of InfoVis as a whole.
• Image 3.1.2 :
  
• Insight :
  The image above is the result on clicking on the first highlighted article published by Robertson in 1989. It displays the result of clicking on this first article, revealing that Dr. Robertson's first initial work in Information Visualization was in the areas of Visual Design and UI. Further interaction with the search results (shown below as Image 3.2.1) we find that the most recent highlighed article, published by Cockburn and McKenzie in 2001, builds on Robertson's Data Mountain, as indicated in the abstract. In actuality, exploring all the references to the Cockburn and McKenzie article, one finds that all references are previous articles published by Robertson, with the earliest one being "Cone Trees: animated 3D visualizations of hierarchical information." From this, we can conclude that Dr. Robertson has done much work in the field of Text/Docs, the research area of Cockburn and McKenzie's article. Exploring futher, we find that the most recent article by George Robertson is "The Document Lens," published in 1999. This article once again fits into the research area of Text/Docs, further supporting our earlier assumption. Note that "The Document Lens" also deals with Focus + Context and Overview/Detail.
• Image3.1.3:
  
• Insight:
  Given that Virtual Reality is a newer area of research, as compared to the other topics, one may be inclined to find out more about Virtual Reality and the researchers interested in said topic. As mentioned above, we can discover which researchers have contributed to a specific topic simply by clicking on the topic name. In the above image, Virtual Reality has been clicked, returning not only documents that fit into topic, but also the authors/researchers who have published in this topic.

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

• Process :
  We did not focus on this task in particular; however, MonkEllipse does reveal relationships of co-authorship and influence between researchers. Since document details are made available when a document is selected, we can quickly discern a relationship between the co-authors in the article's topic. The published authors list that results from clicking on a topic is another way to identify relationships between researchers. Regardless of whether or not the researchers in that list have co-authored a paper, we can still surmise an intellectual relationship between the authors due to a mutual interest in a particular subtopic of Information Visualization. Finally, the references of a document reveal one author's influential relationship on the author of said document.
  Regarding the conceptual author view mentioned above, this task is another that would be better served by the author view. In this case, we can imagine a visual display of relationships between authors by selecting a single author.
• Image 3.2.1 :
  
• Insight :
  The image above is the result of selecting the most recent article returned by searching Robertson for Task 3.1 above. Revealed in the document details, this article is actually one co-authored by Cockburn and McKenzie. The abstract of this article reveals that the 3D system of interest in this article was "heavily based on Robertson et al's Data Mountain." This in and of itself shows Robertson's influence on this paper. Furthermore, backtracking through the green references shown above, one will find that all three references displayed were authored by Robertson, strengthening the influential relationship of Robertson on Cockburn and McKenzie.
• Image 3.2.2 :
  
• Insight :
  Ultimately, the task of finding the specific relationship between Robertson and Card would be best supported by the author view as discussed above. Despite this, MonkEllipse, as currently implemented, does not support this task well even though some amount of information can still be retrieved. By visually scanning the results of searching first Robertson, and then Card, the user is able to determine which articles overlap in the results. In the image above, the WebBook and WebForager article was relatively easily pinpointed as an overlap. Selecting this document, the relationship of a co-authorship between Robertson and Card is confirmed in the returned author list.
  

OTHER TASKS (optional)

Task 4: Finding details

Task 4.1: Search

• Process:
  Our straightforward interface was advantageous again in supporting searches for specific information. The search field and button are always present in the upper right corner of the interface. When the interactor clicks in the field, it goes blank if "Search Terms" had been there, or keeps his previous search so that he can delete or modify it. Once he clicks the search button or hits "enter," the server is contacted to check for exact or partial matches. All current highlights are cleared except for the currently selected document (if there is one), which remains red. When the results are returned, the titles of documents with matches are highlighted (in green), again with a slightly larger size than their neighbors. The titles can be rolled over to explore the search results and the highlighting remains. When a document is selected, the highlighting is cleared, the clicked document is turned red, and when its references are returned by the server they become green. The document's details are displayed in the bottom pane. This functionality allows searching for specific documents or broad themes.
• Image 4.1:
  
• Insight:
  Above is the document that would be returned to an astronomer who is curious about whether or not any visualization tools have been developed for the field of astronomy. Reading the abstract, the interactor would be pleased to discover that there is, indeed, work being done. However, since only one article was returned, we can see how this particular task would have been monumentally more difficult without the ability to search.

Task 4.2: Document details

• Process:
  Any time an article is selected, the details of that article are displayed in the lower pane of MonkEllipse. In addition to the basic information relevant to a document, such as title, author(s), date, and abstract, there is also a direct link to the selected document in the ACM Digital Library. This lower pane, along with the ACM link allows the user to first determine if this article warrants further exploration by providing the article's abstract. If indeed the article is of interest to the user after reading the abstract, he can then quickly and efficiently access the article via the ACM direct link, as opposed to having to manually search the ACM Digital Library (or any other part of the web).
• Image 4.2:
  
• Insight:
  In the image above, we can see the result of following the ACM Link from the lower pane of MonkEllipse. Returned is another window which contains the user's search results, had they manually searched the ACM Digital Library. From here, the user could then access the full text of the document of interest.

Web Accessibility