Date: 3/4/96
Presented by: Dave Thurman
Topic: Information Visualization 2

The Document Lens -- Robertson and Mackinlay

• In general, the authors have not convinced me how it's different/better than the fisheye technique --
  • Explain Fisheye views better -- how is it that a view that just distorts the image/visualization can result in gaps in the contents of the display?
  • The authors say that using the perspective wall technique is a bad idea because it is unclear how to distort efficiently the corners of the sheet. How do they solve that problem in their trapezoid representation? When they clip the corners isn't there danger of losing information (just like they complain about in the Fisheye view)?
  • At the end, the authors mention that their technique has some similarities to the Sarkar and Brown fisheye technique, but point out that because their's is text-specific (less general) it's faster. Again, I ask, what have they done differently (besides optimization). They make claims in their conclusion about general applications for their technique without showing that it could still be efficient when generalized.
• In the Document Lens, besides using highlighting on search results, how does seeing the rest of the text provide useful context?
• Maybe it was just the bad Xerox, but I couldn't see how the context in Document Lens was readable. Not only was it slanted at odd angles, text was greeked in many cases. How does this allow me to read context better than others? Also, if it requires a 200 MHz SGI, how useful is it really?
• The four diagonal edges of the Document Lens is very unnatural. Is there any distortion method to make the transition smooth?
• The document lens seems, like many others, like a good idea. I have 2 comments/questions:
  • If people structure text in chapters, paragraphs, ... I think there is a reason. Is it possible to incorporate the lens with hierarchical text structure (like multiple plans)?
  • An advantage of having the whole text displayed on screen is to see global structure, search & highlight keywords, .... Is it possible to get that global structure with the deformation given by the lens. (and does the always changing view prevent user from building a model image (static) of which part of the text is located on each part of the screen).
• As well as being used to study documents, does the document lens have any other applications? For example, code or data (as in the table lens)?
• I have found it frustrating to read and scan long text documents on a computer monitor and often prefer to print it out to read it. This is a waste in resources and effort. The document lens seems to approach this problem by giving spatial context and visibility to the area being reviewed. However, does it avoid the annoying discontinuous jumps that can be so frustrating when reading from a monitor?
  editors note: Makes me wonder what about books gives context -- eg the thickness of the two halves of the document you're holding. Wonder if this would be a more intuitive way/could be represented on a monitor.
• Is the document lens scalable to very large documents, such as a book? Is it possible to remain in context once there is so much text that it shrunk beyond recognition?
• How is the transition of document pages layed out in the "Document Lens" system? Is this intuitive for someone trying to find a particular page?
• How useful is the Document Lens for documents with lots of graphics. I am not sure if most documents are inherently 2-D linear.

The Table Lens -- Rao and Card

• The example given by the authors is with a well-known data set that they already have intuitions about. Given the exponential number of manipulations available, what are the chances that one would discover meaningful patterns on a dataset that wasn't so well known? Has anyone researched this?
• How effectively does Table Lens handle multiple viewports? What happens if they overlap? If they are side-by-side, does the border obscure any data/context of the other?
• To effectively utilize table lens, a good sorting strategy seems very important. For textual information an alphabetic sort is hardly helpful, because the labels are not grouped according to their category. Is there any research done on that?
• Table lens looks to me like a mature product, intuitive, easy to use.
• What kind of algorithm is being used to display the table data, so it can be scaled up to huge amounts of data?
• When you move the focus window in the table lens, it must move the visible cells at a different rate than the cells outside of the focus window. Is this disorienting? Could there be a gradual transition between the focus and context windows?
• What advantage does the "Table Lens" have over multiple views on a single table?
• Is the slice and dice mechanism of the Table Lens really useful in co-relating information. The paper mentions that interested people were trying various analysis. How about non-tabular data?

Questions relating to both papers

• Can't the Table Lens method be used very successfully to view documents as described in the Document Lens paper - as the documents are laid out in a 2D rectangular array - which is the same as a table of rows & columns?
• Would it be useful to provide an abstracting mechanism in the Document Lens (or Table Lens) that either automatically or through user specification hierarchically groups pages into sections, chapters, etc. ? If this was done, would it be possible to remain in context while examining an area close up?