Like most normal people, I was just having an interesting telephone conversation with a friend of mine (at 2am) about Google, Yahoo, Friendster, on-line marketplaces, approximate searches, and some secret stuff. Along the way I got to thinking about some of the fundamental similarities between Google (those who mapped the relationships among web pages and put them to use) and Friendster (those trying to map human relationships and put them to use).
It occurred to me that Friendster needs FriendRank. Like Google's similarly named dead technology, PageRank, think of FriendRank as a way of providing a measure of influence among "friend nodes" in a social network. Imagine, for example, that Howard Dean wants to convince me to vote for him. He can either advertise in the hopes of reaching me, or he can be a savvy Internet sorta guy and try to use my social network (thru the Internet, of course) to do the job.
At first you might think okay, that's easy. You just need to find the shortest path thru the network from Howard Dean to me. Then you'd figure out who along the way he needs to contact to try to get to me. Well, maybe. Social networks aren't that simple. They don't always use the shortest path--at least not in the "six degrees of Kevin Bacon" sense. Often times they use the most well lubricated path. Or the path that may result in reaching the greatest number of people who are "close" to me. Or those that have more influence with me in matters of politics, as opposed to something complete unrelated like cat grooming.
You get the idea. Like PageRank, it's a multi-dimensional measure that could prove to be quite powerful if applied properly. It's like a routing problem with different dimensions involved.
FriendRank would quantify that stuff. It's the algorithm used to find paths of social influence in various contexts, for various purposes, and in varying networks. Or maybe it's the value that algorithm produces for a given set of inputs. Either way, it's the idea that counts, right? Initially. Then comes the implementation.
Now, if you go search for references to FriendRank, you'll see a few. So this term (and idea?) isn't exactly original. But some of the real possibilities just clicked for me about 10 minutes ago, and believe me, this example is the tip of the iceberg. Some of the discussion here is really, really missing the point. So try not to get sucked into that void.
(Yes, I'm purposely not saying a lot of what I could yet. It needs time to percolate...)
On a semi-related note, it's too bad there's no Friendster web service API I can use to get the data needed to prototype this, huh? That could be a lot of fun. Or really frustrating, as most hard problems are... :-)
On the other hand, Friendster is not a necessary component in the equation. (Or Tribe.net, or LinkedIn, or...) It's just damned convenient since it's big and centralized. If I could get access to enough IM buddy lists, blogrolls, and so on, it'd be doable but much, much harder.
Okay, bed time now.
Posted by jzawodn at January 02, 2004 02:34 AM
No matter how many times you say it, PageRank is NOT dead.
Besides, I don't want Howard Dean trying to reach me, George W. Bush & Administration has done a hell of a job and he's locked my vote (again).
I dunno. Like pagerank, this seems like it's awful ripe for link bombing and spoofing, if for no other reason than you'd have to do double verification for each "friend" link. (e.g. Are you really a friend or are you simply an "admirer/stalker".)
Heh, Of course I'd also guess it might be interesting to have an "Ememyster" link. It would save alot of time at year end cocktail parties.
It kind of makes you wonder what companies who run these large messaging services do with potential data like this? For example say Microsoft, they run hotmail and MSN messenger both are immensely popular but think of all the underlining data that could be compiled.
Email addresses (to and from) creates a relationship, their email address book, etc. Possibly you could have a factor in the friendrank for how many times they communicate although hard to keep updated unless you are continually monitoring.
I've thought about something similar in the past and I can't imaging how neat it must be to work for some of those CIA/NSA types who are creating relationships to track terrorists. Although I don't find it particularly interesting tracking of keeping tabs on people I think the problem is rather complex. But I am sure people are working on such a system. Just image what could be done with phone records :)
I guess even Yahoo has a IM department maybe you could source some data from them :)
Seems like almost everybody is musing about this topic lately. Just three days ago, Om Malik wrote "My Network, My Way!" (http://gigaom.com/archives/2003/12/my_network_my_way.html)
Solution to Routing
Taking this idea a little further and into the blog realm where you have an abunance of real-time information that is always being updated and is public.. Here is my idea.. say that Jeremy links to 20 of his favorite blogs and possibly 5 of those 20 read Jeremy's blog.
You can create a relationship by one blog linking to another and then back again. See diagram..
Relationship Rules
- Blogger A links to blogger B
- Blogger B links to blogger A (once both link to each other assume the friendliness becomes mutual)
From here you can say that at least they are aware of each other potentially they are an acquaintance.
From here you can start to monitor the bloggers comments on A & B's blogs and see how much "noise" is generated. If a substantial amount of comments are posted by A on B's blogs and B on A's blog you can bump up the weight of their relationship.
From here you should be able to search by the weight of a relation ship to determine the best way to route the request.
One step further
To further refine the routing of communication you could determine the "text context" on each blog and assign it to a category. This would become handy when you have two contacts with equal "friendship weight".
For example say Jeremy always talks about "cat grooming" with blog contact X and politics with blog contact Y. The system could determine the "text context" of your request and search for the correct "friendship weight" and "text context" this would yield a fairly good match and would ensure your cat stylist is not the middleman for political discussions but rather you politically minded friend is.
Implementation
Now say you are a large search engine or have the existing capabilities to mine all data from weblogs at your fingertips (i.e google). You also have the "text context" bit nailed (their ad software).. This should be a fairly simple thing to implement.
things that make you go hmmmm
Obviously some of the privacy concerns come to mind when you start talking about gathering this type of profiling information but it really does make you wonder what companies / governments could do with information.
So for example you implement a system like the one above stated and you compile phone, email, web, medical, criminal, etc into a master profile per person. From this information it would be more than conceivable that someone (for whatever reason.. maybe a bounty hunter or something) would be able to track someone down and know who they *might* have an association to.
I don't think this type of system is live but it makes you wonder.. Data mining I think will be a big industry in the next several years. It seems we are just on the tip of the ice burg. To bad I have no data to mine :)
taking this one step furhter is something I talked with a friend who works at Yahoo and that is taking search one step further, and creating a way of searching other peoples completed searches that have been ranked by that individual. Each individual would then be ranked so you could choose what type of person, links, etc from a Yahoo search page.
Doing this would help in finding items that are not "searched" by normal crawlers, thereby finding true gems on the net (especially when doing research on a particular subject while going through an MBA course, you could search for items related to Coca-Cola's expansion throughout the globe, not just marketing crap about Coca-Cola :) )
very unrefined idea, but one that could be part of the Friendster world (search your friends search resluts/links/bookmarks, etc)
jim
I'm still not convinced that Friendster etc. will amount to anything long term. My contacts, both personally and professionally, are very valuable to me. it's not something I'm going to put out there for public use. I suspect a lot of people feel that way too. Before the search companies start worrying about friendrank they need to fix basic search. As Jeremy has pointed out, its broken again.
Oooh, just imagine how useful that platform would be to spammers.
Now you could get Spam as
From: Anne Important-Friend
Subject: Topic you really like
GROW YOUR P3N1S TODAY!
Rod.
Right on - again - to you brother. The PeopleAggregator WILL have APIs into it - so you could do EXACTLY what you want - with Friendrank.
By being FOAF based, we can leverage all the work the Dean folks are doing, and Typepad membership and Ecademy intellectuals. And we can corral all those rdf semantic web nerds - too!
Here's my post.
Justin is exactly right in his comment above titled "Solution to Routing" where he talks about:
"
Relationship Rules
- Blogger A links to blogger B
- Blogger B links to blogger A (once both link to each other assume the friendliness becomes mutual)
"
In fact, such links are already happening in the wild, in blogs and blogrolls.
This is precisely what the XHTML Friends Network (XFN) incrementally builds upon.
With XFN, Blogger A and B can easily annotate their links to each other with the aspects of their relationship, e.g. whether they are friends or aquaintances, whether they have met, whether they are colleagues or not etc.
Bloggers are already linking to each other, in their posts and blogrolls. XFN enables them take the next logical step and say what those personal hyperlinks mean.
hi Jeremy,
Seems like someone has taken your idea and trying to claim as their own:
http://tech.slashdot.org/tech/08/07/04/140223.shtml
Frank
The social network analysis[SNA] community has been doing this for decades [no, not the ad placement] -- the discovery of who is influential and who he/she influences. Lot of "prior art" in various processes and software in the SNA community.
Here is a simple example using off-line data...
http://www.orgnet.com/KOL.html
Here is another simple example with on-line data [blogs]...
http://www.orgnet.com/NEOBlogosphere2005.pdf
The "mavens" are the influencers who others listen to.
