Blogger Hacks, Categories, Tips & Tricks

Friday, August 25, 2006
Experiments With TagPie
It occurred to me that we haven't seen much innovation on the tag-display front since the heady days of tagclouds. Acting on a whim, I thought I try out a new approach ... TagPie!

Now, pie charts aren't particularly a good fit for this task. Firstly, tag counts rarely "add up to 100%" since there's a huge amount of overlap. Secondly, JavaScript does not lend itself to rendering graphs of any kind - but especially pie charts. Still, I found a free piechart provider (they use a PHP back-end to produce an image) in Gheos for proof-of-concept purposes.

So, it's not clickable, and isn't that flexible. But the concept is there - worth pursuing?

Filed in: , , ,
Posted at 12:44 AM by Greg.
<    >
Blogger Greg said...
What about ... Tag Roulette?

Click to spin the wheel ... where it stops, nobody knows!

<    >
I just want my FreshTags to work with the Blogger Beta. Yes, I'm still whining.

<    >
Blogger Johan Sundström said...
It makes perfect sense for mutually exclusive tags a k a categories (at least I think that's the more-or-less consensus about "categories" semantics on the web). It is more or less doable client side too if you use coming standards like SVG, or Canvas. With a bit of luck we'll see either or both in IE7, I'm hoping, so it's for the commons too.

<    >
Blogger Johan Sundström said...
You might like to try playing a bit with an earlier SVG pie chart example of mine and Jeff Schiller's -- it's not overly difficult, and while a bit geared towards showing browser statistics, you shouldn't have an overly difficult time rewiring it with tag names only.

<    >
Blogger Johan Sundström said...

<    >
Blogger Greg said...
Thanks, Johan!

Very cool stuff - I saw some pages on Canvas during my (brief) search around. But this SVG certainly looks the business!

Goes okay in FireFox ... just waiting on IE.

Still, is it a reasonable way to present tags? Given the possibilities of SVG, I wonder if something else couldn't be whipped up.

<    >
Blogger John said...
What if the chart was clickable, to show the titles and teasers for the 5 most recent posts within each slice o' the tag-pie? (in the same window, of course ;-)

Is this also an argument for 2-tier tags, or for the pie to only pick up your 10 most frequently / recently used tags? That way the chart reflects the bulk of the content (more like categories) & the extraneous tags that are there for searches get ignored...

Should the web allow for rel="maintag" (or similar) as well as rel="tag"?

<    >
Blogger Greg said...
Yeah, I think clickability (and interactivity more generally) is indeed important. Johan's pointers to his own geeky experiments with SVG suggests that in the next year or so we're going to see interactivity in browser widgets on a level presently known only to Flash.

I was thinking along the lines that mousing over a "tag sector" would show you posts with that tag, while clicking on a sector redraws the pie with the breakdown for posts with that tag ie drill down. It would be a graphical way of showing the "related tags" column in the delicious interface.

I like the idea of the most-frequent tags acting as de facto categories (ie a mutually exclusive subset of your tags). OK, on a roll here ...

Rather than making this explicit through a "maintag" construct, I'd rather see this happen in a "fuzzy" (statistical) way. A simple (though time-consuming) algorithm exists that could determine the most "category-ish" tags in your set. Eg for Freshblog, it might turn out that the set:

{feeds,, how-to, search, culture, inbounds}

constitute "rough categories", in that 90+% of posts contain exactly one of them (ie less than 10% of posts "miss out" or have "overlapping" categories.)

This would also mean that your tags might get "promoted" to categories over time, or work as categories under certain conditions.

Eg when you drill-down on Freshblog to "bookmarking" then the set
{delicious, simpy, netvouz} might constitute "rough categories", even though they wouldn't make the cut across the whole Freshblog tagspace.

Maybe we need a Ning app for rough category extraction? Might involve hammering the delicious or Flickr webservices a bit too much though - perhaps it could be done through sampling?

Any hard-core computer scientists out there want to have a stab at efficient algorithm design? It seems like a probabilistic version of 3-SAT - does that make the problem NP-Complete?

eXTReMe Tracker