The Relationship Test | Hacker News Reader

The Relationship Test

dataclysm.org

64 pointsolugbam54 comments

I'm getting this response from facebook:

{ "error": { "message": "(#4) Application request limit reached", "type": "OAuthException", "code": 4 } }

Looks like they've drilled facebook a bit too hard. Any chance Christian Rudder will open source this so we can run with our own app tokens?

Shame - I was curious how this would compare to facebook's internal friend ranking (discoverable through methods such as this: http://thenextweb.com/facebook/2013/11/10/script-shows-faceb...)

This is the (jsbeautified) source of the script from your article (http://pastebin.com/kvvYXKKX). I'll wager grammar_costs is just a metric used by graph search and not your friends' coefficients.

Hmm, I'm not entirely sure I understand the "assimilation score". My partner Sarah is at the top, which makes sense, but then some people I barely know (brother of an ex, etc) are like #3 and 4, even though we only have two mutual friends. I have approximately 1,000 friends on FB... so maybe that's broken something?

Same here. Top three are partner, mother and sister. Makes sense. 4 is someone I haven't talked to in 8 years and barely know.

Similar questions here, and I have much fewer Facebook friends (around 350). My wife is #1, and it's not even close (66560 compared to #2, who is 8145).

But after that, I have a few people I don't care at all about in my top 10 (I'm thinking of just removing 3-4 of them now that I see that they're still my friends), and a couple of my closest friends have a score of 1.

From the linked paper [1], it looks like the scores are calculated based on a concept of "dispersion", which is how closely a particular individual in your network is connected to each of the separate clusters of people you know.

Each of the clusters in the graph represents a "social foci", such as co-workers, high school classmates, college classmates, etc.

The dispersion score calculates how well someone you know is connected to these multiple disjoint clusters of people in your life, and tries to show that the person with the highest score is most likely to be your romantic partner.

This probably doesn't work so well in trying to identify friends, since it's likely that a close friend from college may not know any of your friends from work, or your high school classmates.

1 - http://arxiv.org/pdf/1310.6753v1.pdf

I think this had a field day with my disparate groups of ~300 friends/family. I have two huge clusters of school vs work, with school split into high school vs college, and 4 smaller, fuzzier clusters of other groups (gamers, irc buddies, 2x family). My husband came in first with a score of 1310000+ with our roommate second at 343000+.

Most of the next 20 people or so (scores >30k) are people we invited to our wedding that we found out later friended others that attended our wedding, or are people part of multiple groups (tech people that I game with). I'm not sure I'd call these people more central to my life just because they're more social though, after a point.

Number 1 is my brother, which makes sense. He knows many different groups of my friends.

Number 2 is a friend who's a movie producer, ie a professional friendmaker. He's got a zillion friends.

After that it's kind of random.

Three of the women I've actually dated are literally at the bottom of the list. Makes sense, several times I've dated completely out of my circles. Funny to see though.

What's weird is that one of those women's friends, who I otherwise don't know at all, is near the top. Well above the woman I actually dated. We have no other friends in common. Interesting.

From a quick skim of the algorithm paper, here's what I understand of how it picks "high assimilation score" people:

1) look at all links to mutual friends; categorize friends based on their connectedness to different clusters (ie, "school friends", "work friends", etc.) 2) look for friends who are mutual friends with people from multiple clusters (people who know both your family and your coworkers -- "bridge friends") 3) look especially for friends who are friends with multiple people from category 2 (people who know most of your bridge friends)

The assumption seems to be that people who are peripheral friends will only know others from one circle. People who are somewhat close will know others from multiple circles. And people who are the closest to you will know a lot of your multi-circle friends.

(FWIW: this put my wife well ahead of anyone else, but the rest of the ordering seems barely better than random. A guy I played video games with a couple times back in 1998 is #2 by a large margin; the guy from the same video game who my wife and I hang out with several times a week and who co-runs some major projects with us scores less than 1/4 of that. My dad clocks in just behind the new youth pastor at my church who I've known for all of two weeks.)

I think it makes sense that your wife (or SO) would score very highly; if someone is close to you, you typically introduce them to friends, family, co-workers.

A more interesting question would be to ask if it has any predictive power.

Though, also, FWIW, I can't get it to run (getting an error already reported in another comment).

Yeah it seems like it is also a bit zero sum - the higher your 1st match is the lower the rest will be.

This probably happens because if your partner has a large number of connections to the other important people in your life, your partner (/person with most connections to all the various groups) then absorbs all of those people, leaving only people with no connection to those groups to take up the runner up spots.

'Inanimate carbon rod', a page that was a friend's joke and has never had any posts, came 8th on my list.

Well, that's a Simpsons joke[1], so maybe it's popular because of that.

[1] http://simpsons.wikia.com/wiki/Inanimate_carbon_rod

Hmm, I'd take it with a grain of salt, without knowing how the score is calculated. All of my closest friends are ranked way, way down the listings.

I do see four distinct clumpings, with very little overlap: my large, extended family, the people I went to high school with, people I knew in college, and the small group of coworkers I have now. Which is kind of interesting, but also makes total sense, since those are completely disjoint sets of people.

"it will show you ... which of your friends are most mathematically important to your life."

Interesting use of the term "mathematically" to cover the likely difference between the model and reality.

And the only purpose in participating, as far as I can tell, would be to understand how close their model can come -- by definition, I almost certainly know who my important relations are.

#1 is my wife, by literally an order of magnitude. That's about right; we have a truly extraordinary relationship.

After that, unfortunately, this ranking is nonsense. #2 is someone I met once and have never communicated with since. We have only one friend in common. Totally bizarre. My good friends are scattered between ranks 5 and 50, amidst a sea of near-strangers.

Exactly the same here. #1 is my wife and #2 is a relative stranger, which I should actually not have on Facebook. Probably a bug.