Game Theory in Viral Content Propagation
Given the popularity of networking sites like Reddit, Digg, and Facebook, it’s extremely interesting that researchers still don’t have a firm grasp on how exactly content can spread so quickly. While the use of the word “viral” is often used, researchers Andrea Montanari and Amin Saberi have found that content may spread along lines determined by game theory rather than an epidemic.
To elaborate, often popularity growth is often described as epidemic: nodes that have a high clustering coefficient, like a densely packed group of friends or large business network, increase exposure and then adoption. This then spreads to the links that span between groups with a weak, local bridge. However, location and density are not the only factors in the spread. The researchers argue that in some cases, adoption of a technology or new game is often determined by how many and which friends of yours adopt the technology. They demonstrated this via theoretical scenarios in which they create a connected network similar to friends in a social group. Each round, they calculate whether each node has decided to adopt the technology based on their surrounding nodes’ behaviors. While the results were weighted to favor imitation of neighboring behavior, to add noise their results, the researchers added nodes with incomplete information, e.g. not everyone is aware of everyone else’s behavior. They also added the behavior that nodes with no edges would make the decision to adopt the technology on their own, similar to people who are uninfluenced by others.
The most interesting aspect of this study were the final effects of running these simulations. Instead of behaving like an epidemic where highly-connected localized networks facilitate large amounts of adoption, they found that these networks turned into roadblocks because of the added pressure. In other words, because people had so many friends, they had many conflicting friends and so a unilateral decision was difficult to come to. To adopt a technology, the individual nodes had to have many friends adopt the technology previously or be ignorant of its neighbors.
Additionally, the simulations indicated that more local and less-integrated networks facilitated adoption more quickly. Nodes with fewer ties found it easier to adopt something whereas nodes with many ties actually slowed down the rate of adoption in some instances. My interpretation of these findings is similar to normal social interactions. On a large scale, it’s hard to adopt new behaviors you’re unaccustomed to like suddenly starting a sport or playing an instrument in an environment where others have already done or not done so. On a smaller scale, say your more nuclear friends and family, it’s much easier to start one or the other because of difference in social pressures and expectations.
This study directly shows why research on networks is necessary. While the data on large networks is available, we are still exploring the mechanics of information propagation in our networks. We see that while networking can often be predicted, such as business communications correlating to positions of power, the local workings can show new behavior not found on a large scale.
Articles referenced:
http://arstechnica.com/science/news/2010/11/game-theory-explains-why-some-content-goes-viral-on-reddit-digg.ars
http://www.pnas.org/content/107/47/20196.full