What is the purpose of the tag cloud in the concept of the specific web-page? They are often used because it is an "in"-thing. This is the question to ask first. The problem described in the article and the questions/propositions might suggest that a tag cloud is not an efficient instrument for the objective. Especially when you think about pagination or search within.

Have you seen this paper about tag clouds? Seems like grouping is a good way to go. nosolousabilidad.com/hassan/im ... [PDF]

I can't think of any examples at all where these tag clouds are the best way to represent the information; rather than scale them, I'd adopt a different method.

I definitely agree that the tag cloud concept is not good for everything, and that it's often used for gimmick purposes. I think it only seriously lends itself to "explorative" approaches, something more guided than search but also something where the user may not have a very clear idea yet of what they're looking for. As such it can accompany the existing navigation, or when the site itself is completely focused on the "exploration" aspect, then it can take this as main navigation concept (e.g. at CoverBrowser.com the main idea was to allow people to freely browse, discover & explore, so I've combined a kind of tag cloud with a search engine). Maybe grouping is indeed the way to go?

I've always thought of tag clouds as a supplemental novelty. More of a dynamic graph than an interface. It would seem to be very inefficient once it becomes to cluttered. If your tags are so vital in your application that you can not exclude those in the lower threshold, it's probably best to drop the tag cloud or implement a better tag organization method in addition to the cloud. I agree that sub-categories could be a better option, but once you are paginating the clouds it's time to rethink the motivation.

I particulary hate it that tag clouds do not lead the eye. It is a jagged zig-zag, *massive* amorphous mass of word in different sizes, where the eyes slide about and can't get a hold/grip on a word. This is at least how I personally perceive them, I can't get any value of them. The cloud aspect does not do anything for me wrt to discovering new things, as I can hardly get a grip on some words in the tag cloud.

The idea of the font-size reflecting the importance of a term is not bad. (Although on coverbrowser.com, the size diffences are too small. You should make some bold statements about what is important and make these word way bigger.)

Maybe you could try a multicolumn approach (= giving the eyes guidance) and in these columns use font-size differences to show importance. Could be a nice experiment combined with some A/B testing or a survey asking visitors to test and compare both front pages.

Graphically, these columns could use a header for each letter of the alphabet.

OK, I created a quick mockup of this idea a shot with the tag cloud on spiegel.de.

Before:

img299.imageshack.us/img299/28 ...

img216.imageshack.us/img216/38 ...

You are contemplating the wrong problem. You should not be trying to scale the cloud up- you should be trying to scale it down. Anything more than, say, 25-30 tags will amost certainly result in the user ignoring this amorphous mass of information.

Use whatever thresholding algorithm is appropriate to your data and users, but, in the end, a smaller cloud will be more useful.

Tag cloud as a navigation element for "vacation rentals site" at least in my case www.rentalio.com showed up to be very bad thing so I drop it.

The only place where I have seen it works well was some news mashup.

I now updated the CoverBrowser.com cloud with sub-categories, by the way.

Tag Cloud can be reduced by grouping the elements to a subcategory .

Hello.

All possibilities to better views at tag clouds are named in the post itself. Sure the application or better implementation and the relation to the field of case is important. Not every application needs a tag cloud. The first days of Opera Community they had no tag cloud. This was the point for me to think about implementing one and its difficulties. Some ideas are named in the introducting words. If you mind having a look, it is here:

th-o.de/_meine/thoblog-kategor ...

you may want to translate it to english, I prefer Google translation service:

translate.google.com/translate ...

greetings, Thomas

Check out static.tocloud.com/unspsc.html and static.tocloud.com/h1b_stats.h ... for some very large clouds.

Couple of ideas I have around the size of clouds are

1) Have a ruler that defines the treshold. As the ruler is adjusted, words hide and show. That is, higher the value set using the ruler, fewer the words on the page and vice versa. This works well, but I found the javascript needed to loop through and do it made it too slow.
2) For multi-dimensional data, like the one in the above h 1b stats page (more specifically at static.tocloud.com/2006h 1btopsponsorsalaries.html) (remove the space between h and 1), the cloud will be by the first dimension and clicking on the link it will popup another cloud for the second dimension.