delirium happy

Just keep on trying till you run out of cake

Previous Entry Share Next Entry
delirium happy
I recently finished reading Diamond Age by Neal Stephenson. In typical Stephenson fashion, this is an excellent book, with many interesting themes and ideas, and something of a damp squib ending. One of its ideas is that in a future world, with global telecommunications well advanced beyond their current levels, the current nation states based on geographic boundaries could not hold. Tax collecting would become next to impossible due to inability to track transnational purchases, for instance. And as more communications took place via technological, long-distance means, more meaningful groupings could be made up of those of a similar mindset, rather than those of a given residence or birth place. And this set me off thinking about various social groupings.

Most online groups today -- be they on Usenet, LiveJournal, web bulletin boards, IRC, or anywhere else -- are either extensions of real life groups, or else they are based upon a single comonality or shared interest. And while I have certainly come across people who think that just because two people are both transgender/Terry Pratchett fans/LiveJournal support volunteers/people with 8 letter long surnames/whatever that they must automatically be soulmates, I think that most people with a modicum of intelligence would agree that this isn't the case. In general, one shared interest equates to nothing more than a single shared interest, which can serve as a nice ice breaker to start up conversations, but little more than that.

So how then do we go about creating more meaningful groupings? Social networking sites such as friendster and orkut seem to be in vogue at the moment, but they strike me as horribly naive in many ways (and I'll readily admit that that's a view formed off scant evidence, as I've not really properly looked at either). The friend of my friend is not necessarily my friend. The recent "most trusted non-friends" meme that's been going around LiveJournal is a good example of this. Most of the people listed were people that I deliberately avoided adding because I don't like or don't trust them. Or, to put it in more contract terms, most people have many different interests and beliefs. They are likely to be friends with people who have overlapping interests and beliefs, but there is no guarantee that two different people who overlap with this person will also overlap with each other. And we also need to remember that sixdegrees tried the same thing years ago and folded because everyone realised they were crap (though it appears that they may be coming back). I generally don't think much of that sort of site, and certainly don't see it as any sort of way forward.

So, what then? I was thinking something akin to LiveJournal interests, only more sophisticated. Obviously, something simplistic such as checking the number of common interests is doomed to failure. Equally doomed to failure is something along the lines of the old LiveJournal findsim algorithm, which took into account not only the number of shared interests, but also how many people listed each one. Consider for a second, two people who both list "Terry Pratchett", "Douglas Adams", "Neil Gaiman" and "Red Dwarf" as interests. Then consider two people who both list "bisexuality" "computers" "mythology" and "jazz". The latter group are much more generic, and so probably have more people who are interested in them, but I'd suggest that that pair would probably have more in common than the first pair, who really only have a shared single interest listed 4 times.

What you have to do, I would posit, is compare how likely the shared interests are to occur together. So, score points for each common interest based on how rare the interest is, and then points for the intersections of interests based on their rarities, and so on. I'm sure someone smart could come up with some sort of nice algorithm to use for it. I could probably cook up something vaguely appropriate myself if I had the motivation. You could also have feedback as to how accurately it worked, and then incorporate that feedback into the algorithm.

The other problem with things of that ilk is how to break the ice. I know that in situations like that, I would tend to see someone, think "they look nifty" and then promptly proceed to never communicate with them. I would imagine that a lot of people do likewise. I'm not sure how I'd go about trying to solve that sort of problem. Possibly have dynamically auto-generated bulletin board type things and auto-subscribe (wothon the confines of a site) people to it if they're deemed to have sufficiently close interests. Or maybe have people be online and say "find me someone to talk to" and the system would look through everyone else online saying the same thing, find someone with compatible interests, and then give you both the other's name. Much less awkward that way, and possibly workable. I don't know.

And of course, something of that ilk could never be done on LJ itself, both because of the massive server load and change of focus it would require, and because the LiveJournal interest list just isn't remotely sophisticated enough for it. I certainly would be interested to see a site that did try something like that though. (this is the point where someone tells me that it already exists and is crap).

  • 1
A lot of dating sites do /kind of/ what you've proposed. for example makes users fill out reams of interests that you can match with other users on the site. I think LJ Match does that too, in a way. On LJ Match you can find other LiveJournal users who sign up for LJ Match by comparing interests and ranking most similar. Like I said, /kind of/.

Interesting thoughts.

I think it's important to note though that someone who notes four facets of one interest probably has that as a more defining interest for them than someone who only lists one aspect of it.

So the person just listing jazz and just listing mythology may not have a deep interest in them.

Because each interest is not equally important to a person or for compatibility. And I think matching on most important to the person interests is key. Kind of the way I can deal better with someone who is fairly liberal but has vastly different interests than someone who has many shared interests with me but feels that any medical care is sinful and against a deity's wishes and everyone should marry someone of the opposite sex before they turn 21 and then have whatever children they have.

I'm not sure why the LiveJournal interest list couldn't be used for this. It should be possible to aggregate information from individual's users lists to find out which interests are likely to be "compatiable" (for example, most users interested in "latex" are also interested in "rubber", from which we can deduce that people interested in one are likely to be interested in the other).

Server load issues aside, it might be interesting for a site which is both the community-meeting thing you propose, and a blog site, to have a "interesting" list which looks like a friends page and contains recent posts from people whom the system thinks you'd be interested in...

  • 1

Log in

No account? Create an account