Glossy pictures and diagrams

on

Valentine, love, hate: Twitter VennIn the spirit of Many Eyes, Jeff Clark has been developing visualizations of various kinds, including those of various Twitter collections. For example, his Twitter Venn diagram looks at intersections of tweets with three user-specified terms to help understand something about the way different concepts co-occur. Other visualizations look at word distributions associated with pairs of terms, and term use timlines.

The graphs are pretty and, perhaps, informative. His goal is to visualize complex data that don’t lend themselves to standard bar and pie charts. When these visualizations are effective, they can reveal insight that textual representations fail to convey, but the trick is to understand what is effective when. Tufte‘s design guidelines are a start, but one based on a rather static notion of data visualization. Apparently Bertin was more attuned to interaction, but was still trapped in a static medium.

There are two basic challenges to applying these techniques: deciding which data to visualize, and deciding how to visualize them. The twitter examples suggest to me the possibility of using these kinds of visualizations to support exploratory search, but that shouldn’t surprise those familiar with VOIR. If we give up the notion of treating a search engine as a black box that transmogrifies bags of words into ranked lists of documents, but insist on having it “show its work,” we then gain access to lots of fodder for visual exploration. The includes content such as related terms, facets, etc., and behavioral information that reflects what the searcher (or searchers) have done recently in pursuit of the evolving information needs.

How to display such data is a much more open-ended challenge. We can certainly start with some canned visualizations, but it would be interesting to think about more data-driven approaches to reveal hidden patterns. We (see here and  here) and others (e.g., here and here) tried this in the mid-90s on a small scale, but it would be interesting to revisit this topic in the context of modern data sets and computational capabilities. The goal I have in mind is not just graphic representations of single quantities related to information seeking (e.g.,  Tweeder) or overviews of collections (e.g., InfoCrystal) but actually synthesizing interactive visualizations of rich, multi-dimensional data that was either retrieved in an information exploration session, or that characterizes users’ behavior while engaged in such activity. Both kinds of visualizations should help users understand and reflect on the complex activities characteristic of HCIR.

We’ve got a long way to go toward being as facile with graphical means of conveying complex information as we are with textual means, but it is heartening to see efforts such as Jeff Clark’s in this direction.