Re: LLD cloud (with a specific attn to William!)

On 9/9/11 6:44 PM, Tom Baker wrote:
> On Fri, Sep 09, 2011 at 12:12:11AM +0200, Antoine Isaac wrote:
>> Yesterday, I've looked at
>> http://semantic.ckan.net/group/?group=http://ckan.net/group/lld and was
>> amazed by how crowded current status. It seems like william's graph is
>> becoming victim of the LLD success :-)
>>
>> After a while (several minutes, I think), the moving picture stabilize, and
>> it's possible to change it. But that's really tedious, and it doesn't improve
>> the readability much (see attached screenshot). I also can't reduce much the
>> gap between the "real library" part and the RKB-explorer e-prints cloud...
>>
>> I was wondering whether you knew a solution to play with the style of the
>> graph, so that we could have a slightly better picture. Thinking especially
>> of changing: - the size of font (or put the names of datasets in bold) - the
>> distance between nodes (is it the "repulsion force"?) - centering the labels
>> on the nodes
>>
>> Of course I'd understand that this is not possible, given that William is
>> just using an available library. But in case anyone would know an easy way to
>> change it, I'm asking :-)
>
> If there is an easy way to improve it (and I have no idea what parameters can
> be juggled), then we should do it, but I don't think we should sweat this one
> too much.  There are really just four places where the headings are unreadable
> because they overlap.  If the graph could be tweaked to fix just those four
> spots, I'd consider it "good enough" for our purposes.  In our report, as I see
> it, this graph functions more as an "icon" than a source of information -- an
> impression of what one finds when clicking on the link, which is provided
> prominently just below the graphic.
>
> In fact, I'd be inclined to take the image you attach "as is" -- because we
> don't have the time to improve it, or to be more precise are not sure how we
> can do so -- even with its overlapping labels.  Rather, the accompanying text
> should explain in half a sentence how the graphic is generated automatically,
> on the basis of an algorithm, and that it represents the state of links at a
> certain point in time -- i.e., a snapshot that has already changed
> significantly since our earlier drafts and will surely look different three
> months from now.  In my opinion, it is more important to make that point than
> to make the graphic itself entirely readable.  In fact, the text can comment on
> the difficulty of rendering a complex and evolving web of links in two
> dimensions, readable, given the current explosive growth. The overlapping
> labels, then, are almost not a bug but a feature.  The link to the CKAN graph
> generator, then, should not be labeled with "Original at:", but something like
> "For an updated snapshot, see:", because what one sees when clicking on the
> link is not going to look "the original" of what one sees in the report.
>
> With regard to the graphic you attached (if we use it "as is"), it might be
> worth commenting on what appear to be two distinct clusters: one for linked
> bibliographic data and one for linked authority data.
>
> Tom
>
> [1] http://semantic.ckan.net/group/?group=http://ckan.net/group/lld
>


Tom,

You have better eyes than mine, there are certainly more than 4 places where I can't see much ;-)
And yes, I agree, I did not want to make this improvement a mandatory thing: just ask whether anyone knew of an easy fix.

Thanks also for the suggestions for the "legend". We've still have to write a small paragraph of intro for that figure, and you're preparing it well :-)

Finally, the cluster effect comes from the fact that RKB contribute dozens of datasets, one for each part of their (big) database. I have used that knowledge to drag many nodes away from the center by just moving one of them, making the rest more readable.
If you feel that the screenshot I have sent requires explaining this, then I think I prefer using a less readable but more honest pictures, like the one attached now :-)
In fact an easy way to make the graph better would be to replace all RKB datasets by just one big bubble...

Cheers,

Antoine