Sound off on tagging here. I've pasted excerpts from a recent article on Flickr, Metafilter, del.icio.us, etc.
It seems the question will be whether or not emergent relationships/categories will be more useful than static ones. What's lost in the precision and definition might be gained from the social. That said, social value add is often as much about trend/awareness/attention thresholds as it is about anything intrinsic to the information.
"....Though many Web sites have long embedded search keywords, or metadata, tagging has a social component that gives it its power.
...
Warns Danny Sullivan, editor of the online newsletter Search Engine Watch: "The noise and deliberate manipulation will probably just bring the system into a crashing halt....."
www.cnn.com/2005/TECH/in...ap/index.html
It seems the question will be whether or not emergent relationships/categories will be more useful than static ones. What's lost in the precision and definition might be gained from the social. That said, social value add is often as much about trend/awareness/attention thresholds as it is about anything intrinsic to the information.
"....Though many Web sites have long embedded search keywords, or metadata, tagging has a social component that gives it its power.
...
Warns Danny Sullivan, editor of the online newsletter Search Engine Watch: "The noise and deliberate manipulation will probably just bring the system into a crashing halt....."
www.cnn.com/2005/TECH/in...ap/index.html
posted by:
|
SF Bay Area
|
-
Re: Personal Search and Tagging
Thu, May 5, 2005 - 10:33 AMhrm... i thought i replied to this last nite... very strange.
great article, Adrian. i think it mentioned just about every tagging system out there :) here's where i see the greatest challenge:
"Tagging is something selfishly useful. It helps you understand and categorize something for yourself" -- David Sifry (Technorati)
"I don't think in the context of Flickr that there are bad tags, the point is not for you to find all of and only pictures of elephants but to give people a few extra tools to organize their own stuff." -- Stewart Butterfield (Flickr)
we're creating different languages with our tagging. those languages are personal which is why i believe they are useful but they're also incompatible at some level. so they're very useful at the personal level but how can we use them better in a global sense? can we use them as a building block for the Semantic Web (or some other animal like it)?
what we have right now, is mostly algorithms based on co-citation. people tag porche with car and automobile so those tags are probably related. in delicious you get related tags that are probably based on different people tagging the same URL with diferent tags. these are good methods but they're somewhat limited. how can we do it better? i dont really know.
My friend Ryan ( theryanking.com ) suggested (in the case of systems like delicious) using something like TF-IDF analysis of the tagged document to give meaning to the tag itself. this would have to be done per-person (the basic assumption is that every person uses a different vocabulary).
i'm thinking of extending that with something similar to automatic translation algorithms but this is really just a thought no even an idea yet. i'll post more if i ever flesh it out into something more substantial.
-
-
Re: Personal Search and Tagging
Fri, May 6, 2005 - 12:55 AMI am currently exploring the more practical part of the subject. I have build a system for searching the content of pages from loaded bookmarks.
My line of thought was following - when you need to find a page you have allready seen you must remember something from that page and you can readily use that remembered information, which usually will include some characteristic words or sentences, for the search. Thus with search humans need to make least of mental work for retrieving some allready seen information. And in fact you don't need any other mechanism.
But the first user at once started to complain that he needs his directories of bookmarks. I had to find why and here is my hypothesis: with directories (and tags) the user at first glance sees what are the options, he is presented with all the files and subdirs on the current level (or all tags), thus he can choose automatically without thinking (a habit can develope quickly) while with searching he needs to think about what are the possible search terms. The difference is similar to that with Instant Messaging with Presence and email, with Presence you see at once who is available just like with directories you see at once what you can search for.
So the conclusion is that for a complete system you need both mechanisms. -
-
Re: Personal Search and Tagging
Sun, May 8, 2005 - 4:16 AMZbigniew,
your user's comments remind me of the old days of collaborative filtering. some friends of mine were asking friends to upload bookmarks, to get voluntary (permission based in today's lang) associations. That was 10 years ago!
it strikes me that you might have to choose at some point to emphasize one of two things if you're going to do either of them well. Either:
a) making the most of organic, individual directory structures (each of us has his/her own personal ways of organizing, and most of us are subject to subjective decision making... to wit, less than ideal organizing skills/knowledge). the goal here would be "better" directories, or better, and living (dynamic, updated) associations
or
b) pulling the meta relationships out of the associations you can glean from a crowd of submitted organizations/tags... the goal here would be to manifest trends and associations that reflect communication, or the social relations extrinsic to the data (as opposed to the relations intrinsic to the data). to wit, popularity.
granted, (b) gives you a less accurate set, but one that beats conventional taxonomies in that it's living (ok, it's auto-poetic). it changes w/ the trends. it's less appealing to the dbase engineer, but at least it is honest: rather than trying to be more accurate, it's simply diffently organized: by popularity and dynamic selection rather than a reflection of hierarchical expertise...
what does everybody think?
-
-
Re: Personal Search and Tagging
Sun, May 8, 2005 - 6:47 AMMy goal is to start with a) to give users immediate incentives to join and only after having enough users depart in the direction of b).
-
-
-
Re: Personal Search and Tagging
Sun, May 8, 2005 - 4:03 AMthis is all going to get really interesting...
here's something i commented to a tagging blog... i really think we're looking in the wrong place if we just see the directory/tagging/category system, e.g. asking whether or not it's a better way of organizing info. it's about the social -- the relations between people. to wit:
tagsonomy.com/
Clearly there’s a difference between taxonomy and “folksonomy.” (Sorry, going to have to insist on quotation marks with that one for a while. Not sure it’s the right, er, tag…)
The interesting part of the difference though I havent really seen mentioned anywhere. I dont think it’s the actual classifications that result from folksonomic participation (this being the standard distinction between culture and subculture). You say tomato, I say tomato…
What’s interesting is that in a standard taxonomy, categories (associations) are meant to express an intrinsic relation. Folksonomies would, on the other hand, express a social relation. And a social relation ought to make social interests manifest, no?
Shouldn’t a folksonomy thus capture the latent interest among members of a community in this, that, or the other, and render those interests visible? Make what’s subtextual, so to speak, legible.
Tag, who’s it?
Adrian Chan
-
-
Re: Personal Search and Tagging
Mon, May 9, 2005 - 12:41 PM>> Shouldn’t a folksonomy thus capture the latent interest among
>> members of a community
I think you can see that very well in the new interface on delicious - del.icio.us/new/limbo - the little bars on the left of bookmarks show you how popular a link is. It might be interesting to find the person with the most popular links, the guy who represents the zeitgeist of the moment.
I do think, however, that this is a very short range metric. The popular memes on the internet change at an ever growing pace (all your base, anyone?) and so would the most popular tags or search terms (www.google.com/press/zeitgeist.html ). In the longer term, I think that Folksonomies should evolve into a grassroots version of the Semantic Web. Just like the Web, the Semantic Web would evolve from the bottom up (Folksonomies, micro-formats, etc.) not from the top down (RDF and other complex formats, huge databases of formatted data, etc.). looking at tools that can help my tagging vocabulary "break out" of it's personal confines would be a good first step in that direction.
What do I want from a tagging system? here's a partial list. I want to be able to produce and consume tags in my own vocabulary even if they were produced in someone else's. I want to keep the personal nature of my own tags while globalizing everyone else's. I want to be able to connect tags that are related but I don’t want to define those relations. Sometimes I might want to access my tags in a hierarchical manner, other times with full text search, most of the time by tag search. I want to share my tags resources with my friends and colleagues easily and transparently. -
-
Re: Personal Search and Tagging
Thu, May 12, 2005 - 9:55 AMi think you're right on the way in which popularity dynamics shape a collection. which is precisely the point there, i think. to make the collection a reflection of its ongoing self-reflexivity.
i'm with you on tag translation. i was just talking to students the other day about organizing bookshelves. i suggested we do it by color. some of us do it by author, some by genre/topic, etc. If i have a "lacan" section, though, for books by Jacques Lacan, do i put Jakobbsen's book on Lacan alongside? I do, since i dont know names of secondary authors as well.
At what point does your tag vocabulary become ambiguous though? At what point do the associations we make between primary authors and their commentators, or between bands and genres, cease to be clear to others? At some point, what's meaningful to one of us only introduces noise into the system, no? Can that be regulated out by setting some kind of threshold? -
-
Re: Personal Search and Tagging
Thu, May 12, 2005 - 1:16 PMI believe the question on popular dynamics is how do you want to use it.
You can use it for finding interesting links, but then it is only one criteria an algorithm can use for finding them for you.
The www.crummy.com/software/UltraGleeper/ suggests that at some point popularity have negative inpact on 'interestingness' since there is much chance you have allready seen it.
-
Re: Personal Search and Tagging
Thu, May 12, 2005 - 3:16 PM>> At some point, what's meaningful to one of us only introduces
>> noise into the system, no?
that's an interesting observation however with tags adding more doesnt seem like a problem. quite the opposite - if you want to find the more specific resources you use specific tags but if you dont know the tagging vocabulary as well you can use whatever tags you know and hopefully you'll still find the information you need.
as for penalizing popular results, Ryan suggested that idea to me and i dont really like it. the one good argument i heard for this method is that it allows to insert some random new stuff into the system so the system can learn if it's good or not. this is a tempting proposition but i think that learning at the cost of your users getting results you know might be wrong (or bad) is not a good idea. find some other way to filter in new information and always offer the user the best information you can get. -
-
Re: Personal Search and Tagging
Fri, May 13, 2005 - 8:43 AMI think there is more to penalizing popularity over some point than just introducing some randomness:
- what I have already said there is much chance that you have already seen the most popular pages, or at least you heard about it you know where to find it, you don't need the program to find it for you
- there is advantage in being the first to know something - ie in knowing it before it becomes popular -
-
Re: Personal Search and Tagging
Fri, May 13, 2005 - 11:33 AM>> there is much chance that you have already seen the most
>> popular pages
that's probably true in many cases, in other cases completely false - think of the novice. someone's who's new to a field and is doing research to learn the basics. likely the most popular pages on this topic are his best starting points but he has no idea on where to find them and how.
>> there is advantage in being the first to know something
i agree, but you want to know the GOOD stuff. something that's new isn't necesseraly good, interesting or relevant. i think we can do find good and new stuff by looking at the blogosphere. read about iRank here: www.hpl.hp.com/research/i...pers/blogs/ it's a pretty interesting algorithm designed to find "good" bloggers. -
-
Re: Personal Search and Tagging
Sat, May 14, 2005 - 1:10 AMThat's interesting link. I am sure that 'dynamics' is the component missing from current Page Rank algorithms. When I was writing about being the first to discover new ideas I was thinking about a simple algorithm to identify good new ideas - good ideas would gather links faster, so you can kind of predict the popularity of some web page even when it is yet not so popular. But this article seems to treat the subject more systematically. -
-
This is the maximum depth. Additional responses will not be threaded.
Re: Personal Search and Tagging
Sat, May 14, 2005 - 10:47 PMYou're assuming that we can quantify 'good ideas.' Or even recognize them when we see them.
Also, I'm not sure there's really a correlation between an idea being good and popular. I know that crowds have some degree of wisdom, but, as the paper limbo referenced shows, its often a few very influential people who generate and start the process of propagating ideas. It might be more fruitful to find 'people who produce good ideas,' than 'good ideas.'
Anyway, isn't that we do in real life, anyways? -
-
Re: Personal Search and Tagging
Mon, May 16, 2005 - 12:50 AMIf we can't estimage the quality of a web page (an idea) then there is not much to do in the recommendation business. This estimation is our only chance.
As to the correlation than it might be complex, I believe we need some more localized measures of popularity - like 'popularity in the circle of friends'. I think we can do more than just discover the sources of 'good ideas', the sources not allways will produce something interesting for us, we need to filter even the selected sources. -
-
Re: Personal Search and Tagging
Mon, May 16, 2005 - 10:29 AMone of the things our system lets you do is choose your own sources. we can even set it up so you can discover your sources' sources and go directly there...
-
-
-
-
-
-
-
-
-
-
