logical discussion (tags, not trees continued)

A few weeks ago I had a tag-discussion with a very smart chap who's work I respect a lot:

He suggested that filtering should be used to minimise the number of irrelevant tags that will appear if an object/subject is open to tagging by all.
Applying logic to the namespace-creation as it were.

He argued that if "Britney Spears" namespace consisted of a long row of "relevant" tags and somebody adds "nano technology", then that would be silly and should be filtered out.

OK, sounds reasonable... filtering, standards...

But, hang on, say a kid in Brixton started a rumour that Ms Spears had some sort of nano technology implanted in her left breast. Daft of course (and apologies to Ms Spears for the absurd idea), but still, it would not take long before a group in Brixton would use "breast" and "nano technology" as their first tags if they were to find the object of this fine lady.

And, the extra tags will not affect the namespace usability when in anataxonomy finding mode. It would merely broaden the group of users-that-can-find-the-object and perhaps even heighten the knowledge that can be gleaned from the namespace!

The grudge I hold towards tree-structures is the logic-applied-at-source and the ensuing:

• I have to assimilate the logic of the creator (wonder where this web designer placed employees in his navigation tree...)
• I need a manual (PKZ-8763-vy-55? What car part would that be I wonder, looking over the shoulder of my friendly Audi repair shop mechanic...)
• I need training (elevating the confusing to "intuitive", but effectively leaving out my not-so-young mother who's quite new to computing...)

With anataxonomy-in-practice we would have none of that, so why apply any kind of logic to free-tagging? Then the whole replace-tree-structure-exercise would be for the birds...

Let's not.

Let's have logic-free anataxonomy and bye-bye to training, manuals and loosing our ways :)

Comments

[Steve Cooper]

If you have public tagging, why not do something like reference counting or votes on tags?

Say 99 people mark an article with bitney_spears. One person marks it nano_technology. You might create PageRank-like statistics for that article, making it '99% britney' and '1% nanotech'.

Then, in search results, you can order by PageRank. Things that everyone marks as britney float to the top of britney searches, the same for nanotech, and a search for "britney_spears AND nano_technology" would return your breast-rumour page.

Errant tags would drop off the bottom of search results with time, as more and more people tag it with the same tags, and fewer with unique, probably noisy tags.

[sig]

Steve,

excellent idea!
Increased usability, but will also fit nicely with the effort to enhance the value and knowledge built into the "namespace"... thanks :)

Heh, if we could pinpoint where the users of the tags came from we could even get the idea that something obscure is up in Brixton!

Actually, a twist to that will come with next version when summer vacation time is over...

[Ric]

Steve

That's basically the same logic as letting uni students make their own paths across campus lawns rather than put the paving where the traffic never goes. It works well with Sig's idea of not filtering any tags - with enough taggers, you will get a 'common path' to things, but you still allow for some to find their own way 'across the grass'.

[Ric]

Sig

The trouble with finding the source of the tags is you might learn some things you didn't want to know!

But it is an 'opt-in' thing - if people choose to leave more information (meta-tags) then new things may become apparent. you may get a view on what naturally-forming groups are thinking about - like the Amazon thing; 'people who tagged this as nano-technology also bought ...'

[sig]

Ric, ahh, now we're getting closer to the philosophical and moral issues... or do I smell jokes emerging ;)
(A combination is good, keeps things light!)

Shall I be allowed to see it? Shall I spare myself?

Say somebody tagged my personal "namespace" with something really not nice (heh, as if that would happen :-/ ) - then knowing who did it and why should make it valuable. I may not agree, but hey, maybe they're right?

Only "filter" I can think of is a requirement for the tagger to leave a real link to more information about himself (bloggers only..heh?) as well as a real note as to why (a bit tricky technically, but so what).

Transparency is a wonderful morality supplier - funny how people behave if they cannot hide? That could work I think...

[Ric]

Sig

Serious comment, but light tone. And yes, as I was writing it I did get a flash of a couple of scenarios where the results might not be the most desirable. But I am philosophically inclined towards transparency, and I believe that MOST people will do the "right" thing (dial your own definition) - it's just the lunatic fringe that you may have trouble dealing with. I think you have to be careful not to curtail the creativity of the many to defend against the stupidity of the few; but I'll admit there are some blurry lines drawn in there somewhere.

[Steve Cooper]

> Heh, if we could pinpoint where the users of the tags came from we could even get the idea that something obscure is up in Brixton!

I think this system relies on some concept of a user. At the very least, you need to limit tagging - someone shouldn't be able to add the same tag 1,000 times to increase page-rank. You really want 1,000 users tagging the page.

So - you need users to ensure unique votes. If you've got that concept - say, from signing up with an email address, or an openid (http://openid.net/) - then people could tag -themselves- with stuff.

For example; Sven is a rabid death-metal fan. He tags himself with 'death_metal metal goth thrash'.

Sven marks http://www.smorgasbord.org.uk/, the worlds finest marketing-only death-metal band, as 'excellent'.

Now -- if someone searches for 'excellent AND death_metal', then Sven's tag might throw up that page.

[Steve Cooper]

... forgot to say - this gives you a system that feels a lot like Amazon's "people who bought this also bought..." system. If I'm into the same things as you, I'm more likely to like the things you like.

[sig]

Steve,

I agree to no multiple identical tags, actually the present system sorts that out as using same tag twice is not allowed.

And I also agree that when it comes to public/social tagging; the user should have to identify himself, and add notes as "why this tag was used" - otherwise most of the value of the tag would be lost.

I mentioned it otherwhere - the namespace delivers "knowledge" directly (as in relationship between objects), but relationships are kind-of-moot unless you understand why the relationships exist (even with banal stuff like the nano-technology and breas and Spears example). Students of history knows the issue well, there "source analysis" is of utmost importance...

[sig]

Ahh, forgot: I even would go as far as requiring the user to have information about him/herself available to the users - a blog, a personal site if it's public - a live profile page if he/she is a part of a limited system as an inhouse company system...

Then it would enable the user to glean real information from the tag... as a history buff would do from the library...

[Steve Cooper]

> I even would go as far as requiring the user to have information about him/herself available to the users

Being nice and recursive about this; why not make the user one of the knowledge objects? That is, a 'user bio' is one of the entries in the system, and can be tagged.

* A user tagging his own bio is a self-declaration: I might tag my own bio {programmer c# micro_isv_owner cyclist york uk essex_boy}

* A user tagging another person's bio works like a reputation system. You might tag my bio {verbose_git}

> and add notes as "why this tag was used"

This could slow tagging down significantly, though. If I can just add tags that occur to me, (in a freudian word-association way,) that's easy and results in a lot of tags. If I have to write clever-sounding reasons for each tag, I'll write fewer - probably significantly fewer.

So you have to choose between, say, 20 unjustified tags and 1 justified tag.

There's a principle of psychology (can't remember what it's called ATM) where people want to appear to be clever, kind-hearted, noble, generous, and fair. It's why opinion polls for racist candidates tend to be lower than actual results; when it's you and an interviewer, you want to look your best and are afraid of social censure. When it's just you and the box, you revert to your real self.

Similarly, brainstorming advice is always to avoid critiquing during the brainstorm; people who feel like they might be criticised contribute less.

WRT tagging, if I have to justify tags, those tags will be skewed by my need to appear clever to my peers. I'd certainly never write;

nano_technology: I heard britney had nanotech breasts enhanced by aliens from the planet zorp!

but I might just throw in

nano_technology

as a word-association tag. More tags in the tag soup, or fewer, quality tags?

I'd opt for more, because there's a chicken-and-egg problem. If there are only a few tags in the system, you can't find anything. So the system stagnates. If there are loads of tags, you get more results for searches, (even if some of them are spurious) but you're more likely to use the system, and consequently to tag in your own way.

[sig]

Steve,

going backwards - completely in agreement of opting for more! Let the find/seek method sort the "bloat" out...

The balance between "trying to look your best" and having even a bit of "integrity" will be hard to ascertain. Suspect the "integrity-strong" will have a more important say in such an environment, being the ones who will stand up and tell "why"... as that would give the tag more weight.

But, hey, that's fine as "intergrity" is a major trait for a leader... and leaders already... ehh... lead :)

Making the tagger an object is of course the way to go.

And as we discuss social/public/group tagging the tagger would be very important to the knowledge gleaned from the namespace - so include them!

Perhaps like this:
Tagging requirement: Add yourself with links to accesible information, details and most important - your self-generated tags-based namespace!