How to be a good gossip…or at least a good computational modeler

Here are two things that shouldn’t sur­prise you: Our past expe­ri­ences deter­mine our future behav­iors and our social inter­ac­tions are con­stantly changing. When it comes to humans oper­ating in the world, these are just facts and we don’t think about them that much. But they are deeply con­nected to the way infor­ma­tion spreads.

Right now I’m writing to you and a couple sec­onds ago I was talking to my boss. Later this after­noon I have to write an email to someone on the other side of campus and per­haps my mom will call me some­time this evening. If you were to take a snap­shot of the activity in our social net­work at one point in time:

 Screen Shot 2013-04-03 at 3.34.59 PM

…it will look a lot dif­ferent in just a few seconds:


This isn’t much to get excited about, but it turns out it’s really hard to model this phe­nom­enon com­pu­ta­tion­ally. Up until just a few years ago, net­work sci­en­tists have had to rely on the aggre­gates of these snap­shots to esti­mate the system. Now, with more time-​​stamped data like mobile phone records and Twitter tweets, researchers have the data they need to start mod­eling temporality.

In a paper recently pub­lished in the physics arXiv, researchers from North­eastern pro­fessor Alessandro Vespignani’s lab used phone data from 6 mil­lion indi­vid­uals (about 600 mil­lion calls!) to build a more robust model that not only takes into account tem­po­rality, but also deals with system’s his­tory, which has a huge impact on the way infor­ma­tion travels through it.

I tend to call my fiancé a lot more fre­quently than I call the kid I took to the semi formal in eighth grade (although I still think fondly of you, Johnny Depp). So the con­nec­tion between Rob and me is a lot stronger than the one between Johnny and me. If you don’t con­sider the “memory” of our social inter­ac­tions, the net­work looks a whole lot different…


…than if you do:

 Screen Shot 2013-04-03 at 3.20.48 PM

In the first case there are a lot of lines between the dif­ferent nodes (people), rep­re­senting dozens of paths along which infor­ma­tion can travel. In the second there are only a few lines and a lot of uncon­nected nodes. If you drop a piece of infor­ma­tion into the middle of the net­work (i.e., me trying to spread the rumor that I actu­ally got to go on a date with Johnny Depp in middle school), it’ll have an easier time dif­fusing through the first system than through the second. But the real world looks nothing like the first system. Instead it looks like bil­lions of the second one piled on top of and next to each other.

The colors in these last two images rep­re­sent dif­ferent kinds of people. Blue dots are for people that are “igno­rant” of some piece of infor­ma­tion. The red dots are for people who have heard the story from one person and pass it on to their neigh­bors (gos­sips!). The yellow dots get told the story and then deem it not inter­esting enough to pass it along any­more. As you can see there are a lot more gos­sipers in the unre­al­istic net­work than in the real­istic one. In fact, in the more real­istic net­work, it looks like infor­ma­tion never goes any­where, it keeps get­ting squelched by Debbie Downers who don’t think it’s amaz­ingly awe­some that I got to slow dance with Johnny Depp.

Well, it turns out it’s actu­ally pretty dif­fi­cult to spread rumors through tightly knit sys­tems, according to the paper, of which post doc­toral research asso­ciate Márton Karsai was the lead author. “It’s coun­ter­in­tu­itive because you would think that your stronger ties would help spread infor­ma­tion because you’re com­mu­ni­cating a lot,” he explained. But in fact it isn’t until you tell someone whom you’re only weakly con­nected with that the rumor goes viral, so to speak. That’s because everyone in your tight little net­work already knows the infor­ma­tion (or knows it’s not true), so they don’t feel com­pelled to pass it along. If you tell someone you’re only weakly con­nected with, it’s likely they’ve never heard this story before so they’ll be more likely to tell someone in their own tight little net­work. Then it spreads pretty quickly, spilling from one tight little net­work into the next.

This work was reported in a few main­stream media out­lets over the last week or so, and not sur­pris­ingly: who wouldn’t want to know how to spread infor­ma­tion if not jour­nal­ists? But Karsai said the story is much bigger than the head­lines were giving it credit for (“How Strong Social Ties Hinder the Spread of Rumours”). What’s really cool about this paper is that it’s the first one to ever suc­cess­fully model how memory effects the spreading of infor­ma­tion in human social network.

They’re data showed that a system’s memory, or the degree to which the people are linked, deter­mines how far infor­ma­tion will travel. If you take into account the tem­po­rality of a system (the fact that we’re not all talking to each other all the time), deter­mines how quickly the infor­ma­tion will spread.

But what’s a model without val­i­da­tion? Not much. Remember those 600 mil­lion phone calls they exam­ined? It turns out their models did a much better job mim­ic­king that real world data when they included memory and tem­po­rality in the mix.