Favorite cloud August 22, 2009 5:56 PM   Subscribe

Suggestion:We have a tag cloud - how about a favorites cloud?

Popular favorites are great, but often I'd like to find things that are only moderately popular.
posted by bigmusic to Feature Requests at 5:56 PM (20 comments total)

posted by Pronoiac at 6:20 PM on August 22, 2009 [8 favorites]

We've been sort of collectively geeking out about data-analysis stuff in the Infodump 2.0 thread; it might be worth poking some of the folks there about the kind of information something like this could incorporate and how.

I'm neither here nor there on the idea in general. It might be neat to see. But starting with a one-off analysis just to see what it might look like would probably be a good start.
posted by cortex (staff) at 6:23 PM on August 22, 2009

Pronoiac's mockup makes a good point in a hilarious way. Tags have a textual tag name associated with them, by which they can be grouped together, counted, and displayed in a cloud format.

Favorites, themselves, have none of these things. There isn't anything in a 'favorite' that would make sense to display in a cloud. So instead we'd have to look at the things that favorites apply to and relate to, specifically:
  1. What the favorite applied to (post or comment)
  2. Who did the favoriting
  3. Who wrote the thing the favorite applied to (poster/commenter's name)
  4. Tags on posts that were favorited
I can imagine user names and tags appearing in a cloud format. I can't really see post titles, or entire comments, working in that format.

So, in summary, I'm wondering:
  • What entity you would specifically like to see as the items in a cloud (since favorites themselves would not make sense)?
  • What criteria would you like to see for inclusion and relative size of the items in this cloud?
The Infodump 2.0 thread that cortex linked to does indeed provide plenty of examples of potential answers to these questions.
posted by FishBike at 7:44 PM on August 22, 2009 [1 favorite]

I liked the infodump better when I thought we were dumping all the information and starting fresh.
posted by Eideteker at 8:45 PM on August 22, 2009

With the Infodump alive again, PATCHES WELCOME now applies to shit you can't do with Greasemonkey.
posted by blasdelf at 9:38 PM on August 22, 2009

What would it do, show people's names that have the highest aggregate favorites? Could name it the Snark Cloud.
posted by cj_ at 10:48 PM on August 22, 2009

I'd like a cloud of pony requests which make no sense.
posted by finite at 12:10 AM on August 23, 2009 [2 favorites]

InfoDump + favorites cloud + tag cloud + Harvest Gold background + hugs + more bike threads = hipster.metafilter.com

I can not wait!!!
posted by paulsc at 12:56 AM on August 23, 2009

Clouds are bad. There's nothing a cloud does for you that a list doesn't do better.
posted by Plutor at 1:34 PM on August 23, 2009 [1 favorite]

plutor - so there's no data that's better displayed in 2 dimensions than one? a map of city locations is just a cloud where the similarity metric is distance. I'm not saying clouds aren't overused, but you're throwing the useful data visualization tool out with the annoying flash wankery.
posted by russm at 3:48 PM on August 23, 2009

I think the definition of "cloud" here is pretty clearly "like a tag cloud", where positioning of the individual tags is either arbitrary or simply pointless. In fact, clouds only need to be 2-dimensional because the font's usually too big to fit in one dimension. In a map, positioning has meaning.
posted by Plutor at 7:04 PM on August 23, 2009 [2 favorites]

take a look at something like the Public Whip, displaying voting patterns of UK parliamentarians. this reduces a high-dimensional data set (voting similarity data for MPs) down to a 2-dimensional representation clustering MPs by their voting patterns.

the axes and absolute positions of the MPs are meaningless, but the pattern is meaningful - if it's not a "cloud" then I don't know what is. do you think that a 1-dimensional plot (like a pure left/right scale) would convey as much information? the simplest 1-dimensional collapse would indicate that Clare Short, Phil Wilson, Derek Conway and the rest of the group down the bottom would be likely to vote the same as Alasdair McDonnell, Mark Durkan and the other folks who sit near the middle of the primary labour/tory axis, when this is not at all the case. a 3-dimensional representation would allow even more detail to show through.
posted by russm at 7:59 PM on August 23, 2009

or since you're American, here's something similar with data from the US senate.

I guess the short form counter to "There's nothing a cloud does for you that a list doesn't do better" is "display data with more than one dimension"

posted by russm at 8:05 PM on August 23, 2009

russm: I'm not speaking for Plutor here, but I'll just point out that he stated a dislike of tag clouds, specifically, not scatter plots (such as your US example) or maps or 2d graphs in general.
posted by Pronoiac at 9:01 PM on August 23, 2009

ah, fair enough... I took "cloud" to refer specifically to the tedious flash and java implementations that hide the useful similarity data in a great steaming pile of "why? because we can"...

since when is a page of text a "cloud"? kids these days...
posted by russm at 9:19 PM on August 23, 2009

A "tag network" cloud --- with distance between words related to the likelihood of two tags describing the same thing --- might be interesting to see.

A "tag cloud," an alphabetical list of tags printed in different sizes, is not so useful.

A while ago I made a list of askme questions with popular tags with about 20k links, just to see if I could. Learning enough AJAX to make this a useful interactive tool has been low on my list for a year now. Slight bump now that the infodump is back and richer.
posted by fantabulous timewaster at 8:55 AM on August 24, 2009

Pronoiac: russm: I'm not speaking for Plutor here, but I'll just point out that he stated a dislike of tag clouds, specifically, not scatter plots (such as your US example) or maps or 2d graphs in general.

Yeah, but the geek in me thinks there's awesome potential there.

I'll tell you what I want: a three-dimensional scatter-plot of favorites with cross-references to geographical location, whether or not the poster has a high or low user-number, and whether they have mentioned, say, pants in the last thirty days.
posted by koeselitz at 5:20 PM on August 24, 2009

I read that as "whether or not the poster was high".
posted by cortex (staff) at 5:31 PM on August 24, 2009

I read that as "whether or not the poster was high".

Can we get that information added to the Infodump?
posted by FishBike at 5:40 PM on August 24, 2009

FishBike - without client-side support that'd be hard on an per-post/comment scale, but an analysis of posting history could probably give you a "user is usually high/drunk" tag against the account...
posted by russm at 6:52 PM on August 24, 2009

« Older Denver meetup 10/6/2009   |   I am a fat Special Snowflake Newer »

You are not logged in, either login or create an account to post comments