Online Social Networks Can Be Tipped By Less Than 1% of Their Population

An anonymous reader writes "A new algorithm developed by researchers at West Point seems to break new ground for viral marketing practices in online social networks. Assuming a trend or behavior that spreads in an online social network based on the classic 'tipping' model from sociology (based on the work of Thomas Schelling and Mark Granovetter), the new West Point algorithm can find a set of individuals in the network that can initiate a social cascade – a progressive series of 'tipping' incidents — which leads to everyone in the social network adopting the new behavior. The good news for viral marketers is that this set of individuals is often very small – a sample of the Friendster social network can be influenced when only 0.8% of the initial population is seeded. The trick is finding the seed set. The algorithm is described in a paper to be presented later this summer at the prestigious IEEE ASONAM conference."

Comments

It all makes sense

[Sparticus789]

Is this what Occupy means when they say 1% of the country controls everything?

Re:It all makes sense

[jellomizer]

But which 1%, There are a lot of 1% out there.

The Tea Party is controlled by the Oil Companies 1%.
The Occupy is controlled by the Unions 1%
The Favorite Trend is controlled by the Marketers 1%...

It doesn't seem that you have any decisions to make for yourself, There is always someone else telling you what to think.

Re:It all makes sense

[Baloroth]

That's not really what this statistic is saying. It is using a "tipping" model, where it assumes anyone who has more than 50% of his friends exhibiting the behavior will automatically adopt it. A useful model, but not actually true (like nearly all mathematical models, it is only approximately true in the real world). That means they only have to find a "seed" population to adopt the trend: the model says if all of them adopt it, everyone on the network will. Think of it less like sheep and more like dominoes: you only need to trigger one dominoe to trigger the rest, but that presumes a carefully constructed ideal system. In reality, 99% of people may be sheep, but this study says absolutely nothing about that. It assumes it, rather than proving it.

Re:Wow, Friendster? All 300 Users?

[CohibaVancouver]

Friendster? Wow, you could influence, like, 300 people

American-centrist much, you insensitive clod?

From http://en.wikipedia.org/wiki/Friendster

Since the relaunch of Friendster as a social gaming platform in June 2011, the number of registered users has reached over 115 million. Over 90% of Friendster's traffic came from Asia. The top 10 countries accessing Friendster, according to Alexa, as of May 7, 2009 are the Philippines, Indonesia, Malaysia, Singapore, Thailand, Pakistan, United Arab Emirates, Sudan, South Korea, Bangladesh and India

Sheeple

[Chemisor]

A much more interesting conclusion of this study is that 99.2% of social network users will do anything their friends would tell them to do.

The trick is not just finding the seed set

[timeOday]

"The trick is finding the seed set." No, you still have to influence the seed set, which might be really hard.

Let's say this model predicts that I can end terrorism by converting 100 radical muslims to buddhism. How does that help me? (Simply sending in drones to remove these nodes from the graph, so to speak, will not have the same effect).

Second example, let's say my novel is almost guaranteed to be successful if it gets a glowing review in the New York Times. Well, how hard can that be? Usually trusted nodes are trusted for some reason - because they're reliable. That means they're hard to influence.

"The trick is finding the seed set."

[Lord Byron Eee PC]

"The trick is finding the seed set." No, it's not. The real trick is finding the seed set of the seed set. On Facebook, you have 900 million users. 1% of that is 9 million, which is too large to influence. But 1% of that 1% is just 90,000, something that a targeted advertising campaign might be able to influence.

Re:Wow, Friendster? All 300 Users?

[plover]

Any chance they're just witnessing C&C nodes transmitting spam orders or pagerank gaming links to the remaining 99.2% of Friendster accounts (all of which are hacked and forgotten)?

It's a comp sci paper that is looking for connected nodes in a network, and they're using copies of data sets of social networks as their starting point. They aren't monitoring networks looking for "who is exerting influence over them", they're looking for nodes that are well connected to other nodes, presuming those represent the most valuable people to convince.

Now, could those "friends and families" in the network data actually be there as part of a botnet controller and its zombie minions? Sure, why not? But each one of those would be a single node in the set of nodes as having the right connections. Doesn't mean that marketing to the botherder or the botnet is going to get you much business, but if you were looking for someone who has influence, it would identify the botherder and not the bots themselves.

Algorithm is very simple

[mrops]

I looked at it, and it looks like this

public static boolean willTip(User user) {
  if(user.sex == SexType.FEMALE && user.hotness>(Long.MAX_VALUE-100)) {
      return true
  }
  return false;
}