David,
> Georgi, thanks for playing with it! I've played with your prototype a
> bit, too. Clearly there are ideas to exchange between the two. I'm
glad
> to find someone else with the same interest!
Me too. I think this research area is a very interesting one and I think
it addresses business demand as well.
> > While looking at the publications collection I can select values
from
> the authors facet to filter the publications. Now focusing on the
> authors could be understood as a zoom into on dimension of
> publications, i.e. their authors.
> The user's action of focusing on the authors can be interpreted in two
> ways:
> - trying to filter the publications by temporarily focusing on the
> authors
> - permanently shifting focus to the authors (much like when you click
> on
> a web link, you're shifting focus, you're not temporarily looking at
> another dimension of the current web page)
> The current UI design is meeting the second need, not the first one,
> which leads to the confusion. The "group by" feature is intended to
> address the first need.
I don't think that the web page metaphor applies very well, but I'm not
sure. I think of an explorative query builder, and that's more related
to multidimensional databases. I hesitate to use these terms because
they introduce an huge amount of complexity. But that is the
expressivity I would like to fit into an useable interface. Crazy?
The group-by feature is neat, but it will break as soon as the user's
path becomes as graph. Then the group-by would need a "refocus" feature
as well to enable the user to explore /what/ he could group-by.
And then the distinction between these two ways of filtering is more
confusing than helpful.
Maybe my idea of lifting the path to a graph doesn't work at all...
> > Every filter which is now applied on the authors collection should
> now also affect the publications. An example: Given the task "find
all
> papers from authors in Europe", an
> > user starts looking at papers, selects people -> schools ->
countries
> and filters by "Europe".
> > And thinks he is done. But he is not. Now he has to find his way
back
> again via selecting schools -> people -> papers. That is, in my
> opinion, confusing.
> >
> I understand the confusion very well, actually. You're perfectly right
> that if the task is to find all papers from authors in Europe, using
> the
> link sliding feature (what I call those links on the facet headers)
> instead of using the "group by" feature is not cognitively obvious.
>
> By the way, perhaps it's more interesting for you to try it on the
> "Kennedys" example instead of the "Publications" example. Given that
> there is only one type ("Person"), the link sliding feature might give
> you a different feeling. Try to go from the 1st generation of Kennedys
> (2 people) to their direct children, then to their children's
> spouses,...
You are right, in that example the link sliding works better for me. But
as you says, that's because I'm looking at only one domain (person)
here, and every step is a refinement of the same list of resources. I.e.
the first view shows all people in the dataset.
>
> > So filters should affect all connected collections. Of course, the
> whole thing could be understood as a sequence of filters instead, but
> even then: if I put "papers" in, I should get "papers" out...
> >
> Well, if you think "if I put 'papers' in, I should get 'papers' out"
> then you're still thinking in terms of tables, rather than of graphs.
> With graphs, you can put in "the Kennedys" and get out "the people who
> hired the guys who shot the assassins of the Kennedys". Or if you put
> in
> "publications" you might want to get out "the funders who fund the
> professors at the schools where the graduate students work on topics
> related to the topics of those publications." :-)
Well, I was thinking of the sequence of filters as a kind of pipeline (I
steal that metaphor from my colleague Richard here).
If you think of it as a pipeline, you put in stuff and get a subset of
that same stuff out. Navigating the graph is only done to select
filters, i.e. as you wrote above, the user only temporarily switches
focus.
> There is another reason why the influence goes from left to right
only:
> it's easier to understand when the effect is in one direction. If the
> influence goes both way (i.e., the whole solution is a fixed point),
> then it would be (even more) confusing.
I'm not convinced ;) Seems like a good place for a little user study...
>
> Obviously this prototype will need many more iterations until it makes
> any sense to most users. Please let me know if you have more thoughts
> on
> the prototypes or on the topic in general!
Well, let's imagine a dataset like DBpedia where the user doesn't know
the graph and doesn't know which filters could be applied. A task where
the user has to explore the dataset first. That exploration is much
easier while looking at actual data. And again, the group-by breaks ;)
Cheers,
Georgi
--
Georgi Kobilarov
www.georgikobilarov.com
_______________________________________________
General mailing list
[email protected]
http://simile.mit.edu/mailman/listinfo/general
