First, a general petulant comment. Many of these interface issues are years old. How come we don't use any of them? Is the marketplace just slow, or are people conservative about new interfaces, or are the ideas not quite right? The history of the adoption of the GUI might be useful to consider here, from 19Star77 to 19Mac84 to 19Win95.
I'm all for interactive displays, allowing you to tinker. Bricolage is a useful way to understand a new domain. It's curious that this sort of tool (whether starfields-and-sliders, or other kinds of interactive graphs) are not much more common. Is it just too far removed from traditional publication media?
As for specifics.. the problem with starfields is you have to choose your two axes, and the choice of those axes strongly controls the perception of the data. Allowing you to tinker with the axis choices, and maybe some clever 3d projection interface, might be more powerful. Their use of colour to encode the third axis (movie type) doesn't work.
Tight coupling is an important feedback mechanisms, but only captures exact exclusionary set dynamics. Ie: it only works with hard categories with sharp boundaries. In social visualization my intuition is the boundaries are much fuzzier. So for instance you might say "show me all men", and have a pretty strong dividing line, but you might also just want to say "show me all people in my age group": how do you draw a strict line there?
Pad, An Alternative Approach to the Computer Interface
Pad++:
A Zoomable Graphical Sketchpad For Exploring Alternate Interface Physics
I've always liked the Pad/Pad++ work, I think it's really elegant and beautiful. I believe this stuff has a good shot at actually being used in real systems, although interestingly enough it has limited application so far. The basic metaphor is so clean and compelling, it's immediately obvious how to use it. And yet when you get to nitty-gritty details, the metaphor gets broken in lots of ways.
Infinite scaling is a powerful tool. You can zoom in for more detail, and yet still maintain a view of the context as a whole. I think in general it's more natural and pleasant than other ways of putting lots of scales on the screen, such as fisheye views. But it doesn't work in some ways. The recursive web browser is one example. Pages shrink to infinity very quickly, and the effect as a whole is somewhat weird. For some reason, the directory browser feels more effective - you get stuff placed inside other stuff instead of next to it. Maybe this works because directory folders are a strict containment metaphor, whereas web pages are links.
Dynamic interactivity is crucial in these systems so that you don't get lost moving from one view to another. That puts big demands on the interactive display system - all this rendering is hard work for the CPU! There's a lot of hacking in there to make navigation feel natural.
The space/scale diagram is a very powerful way for understanding how to move about the zoomable space. It's worth paying attention to. One interesting idea is that in Pad currently, you always have a parallel plane of view to the pad surface. But what about one at an oblique angle to the pad, so you get parts of the pad at one zoom factor, other parts at another? Maybe too disorienting?
The infinitiely zoomable canvas is one part of the Pad metaphor, but the other is the idea that the whole world is placed into one 2d space. I feel that this side of the metaphor is less successful and has to be toyed with more often. But can 2d placement really be separated from zoomability? I don't know.
One place the 2d placement breaks is in the idea of portals. I have one giant sheet of magic paper, and yet really I have a bunch of views on the same piece of paper, and some views are nested inside others. It's like I'm reading different pieces of the paper at different magnifications at the same time. Sticky portals only make things worse. This stuff can be a bit disorienting, although it's also quite useful to get real work done. Bookmarks are an interesting alternative, more a set of directions to get somewhere than a representation of that somewhere.
Semantic zooming also partially breaks the metaphor. Things shouldn't change kind just because you view them at a different scale. You have to use semantic zooming carefully so to preserve consistency. And I think that's hard. For instance, I think on the calendar it's a bit weird to have dates appear and disappear in different places. But how else do you do it? Sometimes the semantic viewing can be used to great effect.
Interacting with a Pad document is an interesting thing. You can make regions clickable, but then what's the effect of a click? Sometimes it's activating a tool, changing modes. Most of the time it's a warp, moving you to a new page or view of something. It's important not to warp the user around too much, or else he or she will get lost. An "undo" for motion could be useful.
There are lots of options for applying Pad ideas to social visualization. The simplest are just using a zoomable interface to cram more info on the screen - for instance, I suggested this in my cocktail party visualization of soc.motss.
But Pad is best at displaying information with one conceptual organization, one 2d layout of the info that you then view at different scales. Societies aren't like that, there's no one perfect view. For instance, if you chose to lay out all of Usenet on a pad, with each newsgroup as a box at one scale that you could then zoom in and see the people, then that visualization would really emphasize the individual newsgroup. What you would lose is the identity of people across newsgroups, which might very well be more interesting. To some extent, Pad forces you to choose one layout of information, rather than providing many different choices.
I guess the solution is to use an infinite scaling system like Pad, but allow users to flexibily choose the layout of information themselves. Have different organizations of data depending on what's of interest. But over-configurability is the major sin of user interface design, there has to be some stronger designer view of how things should go. Mixing Pad metaphors with the interactive control of the Starfields paper is worth thinking about.
I wanted to see what a simplistic web crowd visualizaiton would look like, maybe suggesting the capability of the zooming interface. Individual dots are people - location indicates what they're doing, whether looking at one page or moving between pages. Colour would be some sort of information about the viewer (physical location?); the confetti look wasn't intentional, the graphics could obviously be improved.
I think the previous, as a display alone, is pretty hoaky. But combine it with some zooming.. As you get closer to the people, they come into sharper relief and you see their own web browsing histories inside them. Of course, a person's browsing history is itself zoomable, and contains within it pictures of people that he had met as he was browsing those pages. So the picture inside a person is a snapshot of their own life, their own history. You can zoom in on that snapshot, see the people inside someone else's history, and see their histories (at the time they were met), etc.
Toolglass and Magic Lenses: The See-Through Interface
The basic idea of magic lenses is very powerful. However, this paper seems to suffer from feature proliferation. There are too many kinds of lenses described here with too many different semantics. Simplify! One way to do this is to really study composing lenses - I'd rather have three lenses that I can compose than eight different lenses with different functions.
One thing I find confusing is that some lenses are read-only, allowing you to view data in different ways, such as a zoom lens. Other lenses are write-only, changing the way you can affect the view, such as the shape palette. Offhand, it seems to me that the lenses for reading are relatively benign and useful, while the metaphor for the writing lens is very strange.
Some lenses that I'd like to see for social visualization: gender, age, status in community, history within the community, reputation, mode of contribution, ... The lenses themselves can be tunable as well. For example, I might have one "reputation" lens, but be able to tune it to different people's ideas of the reputation of the community members.
One off-hand comment(*), I really like the idea of two handed
interfaces. Why don't we have those? It'd be fairly hard to hack one
up in X windows (but not impossible...), how about in MS Windows?
(*)so to speak.
| Nelson Minar | Created: October 13, 1998 |
| <nelson@media.mit.edu> | Updated: December 15, 1998 |