Jeff Heer just finished a great talk about his work at Tableau Software, where they incorporated history exploration tools for Tableau. As usual with Tableau, there were plenty of impressive demos. One of the interesting points about their implementation is that it is possible to create a meaningful ontology of user actions, and this allows very effective chunking of histories. I would love to do something like that for VisTrails, but I’m not sure how not to require user-defined ontologies, which I think are a bad idea.
I was particularly intrigued by the choice of linearizing the history information. I certainly see the value in saving screen real estate, and that linearization is very effective for this. However, I am not sure that people will adequately perceive the branching ability in a linearized view. It would be interesting to test this hypothesis by comparing session information from Tableau and VisTrails, since we do show the entire history, ugliness and all. If people interact with the system differently, I expect we should see different branching characteristics.
The most interesting piece of information collected during the sessions was that users were about 10 times more likely to undo than to redo. One important question asked was whether this is an inherent user trait, or a consequence of users not typically having access to history. In the same vein, I wonder if undo-as-delete (a good idea that would certainly unclutter some of our history trees), as presented, actually conditioned the users not to try redo, since they quickly learn undone actions disappear.
All in all, an intriguing, thought-provoking talk – go read the paper!