It might take a little bit of time to load if lots of people are trying the demo, so please try to be patient. We spent a lot of time to ensure you'll be impressed.
Feel free to email us at tony or chuck @graphmuse.com if you'd like to discuss further!
That said, the clusters are really impressive. All of my clusters were basically just groups of people from different times/events of my life (highschool people/friends, college friends, colleagues, people I knew when I lived in Syracuse, etc). It's an obvious way to separate my groups of networks, and it was fast. That's useful in so many contexts
What if we were to identify your evangelists based on "cluster completion?" What I mean by this is that we'd look for clusters that go from being densely unregistered friend clusters to densely registered over time, and see which friend(s) were the catalyst(s).
From a purely consumer standpoint, I would use something like this to create my FB lists, if you made that easy. It's also useful when planning events. So, while I would personally never use it to send an app invite, I would use this if I was planning a party, and wanted to quickly see who I should invite and message, etc
Either way, awesome work Tony & co!
Do you view the invites you receive as complete spam? I guess it might depend on the size of your college, and how open the college is to parties, events, etc.
This would be an awesome text-based adventure game, yeah?
I just demoed the app and it's awesome, it basically identified all of my friends into perfect groups, with only a little confusion. Obvious groups like "people I met when I spent a summer at Brown" were 100% accurate and blew me away. High school and college also got their own clearly defined groups. Way to go, I hope other people are as interested as I am.
Are you a Facebook app developer? Are you interested in purposes other than sending invitations?
tony@graphmuse.com
I have some ideas that are worth developing. In fact, we were using GraphMuse originally to determine entire fraternities and sororities. We listed 3,400 Greeks at UPenn in 3 weeks.
Feel free to check out Greekdex.com
or read about it here:
http://www.thedp.com/article/2012/02/new__directory_launched...
As part of your demo, you should let me build a list based on each cluster.
We might include that as a neat little tool in the near future for those interested.
What are your thoughts?
your clusters were perfect lists of family, different groups of friends who knew each other, my wife's family, etc.
Uncaught Error. {"errorCode": 6, "errorStr": "Internal operation failure, unable to reticulate splines!"}
-Do you have more than a small number of Facebook friends? That might be it.
What about a way of creating all your G+ circles based on Facebook friend clusters as well?
Katango basically provided that service for free, but they got acquired by Google. http://techcrunch.com/2011/11/10/google-acquires-katango-the...
Maybe pricing that works out so it is free for general users and has a low price point for "pro" users?
That said - request recipient optimization works, and offering it as a service could be a successful business. I think your primary value proposition (higher CTR's on lower volume) is solid, but would like to see verification that your approach (using clustering) is valid for that use case.
At this point, we're in the process of confirming our hypothesis with case studies.
We did launch this clustering algorithm for our previous venture, Greekdex.com, in which we determined entire fraternities and sororities given just one user's friend graph. Clusters were "completed" over time as more users signed up.
That said, it's obvious that members in fraternities and sororities tell their fellow brothers and sisters about the site that they just signed up for. Therefore, it's difficult to say that our clustering algorithm resulted in more users signing up.
tldr; we're working on confirming our hypothesis with case studies.
What about security? Passing Facebook user access tokens over http opens them up to being sniffed on the wire, which would allow anyone to impersonate a user of your app. Maybe ssl is supported, but all of the api documentation examples use http.
We're also in the process of determining our pricing scheme, which is why that is not detailed on the site yet.
Facebook provides you the user's list of friends, and each friend's list of mutual friends with the user. From there, we construct the friend graph, and then run our clustering algorithm on it.
You're right about Facebook not providing friends of friends. Supposedly they used to back in the day, but not anymore.
Great service, did a great job on my list of friends. I would love to use this, but having to send over user access tokens is a little scary, even over HTTPS. Have you guys considered licensing this? Would be cool as a heroku plugin...
I don't quite follow.
Arbor is pretty slick, but we haven't tested it out extensively. We have our eyes on d3.js and three.js as well.
