Global Information Flows

Short post this. Yesterday saw the demise of my laptop c/o Windows Update and I’m utterly locked out. A replacement machine has appeared but I’m still in the process of recovering a whole term of work …

IntelDump alerts us to a new report by the National Academy of Sciences, “Tracing information flow on a global scale using Internet chain-letter data”. This “breaks the popular misconception that information online is spread widely in very few steps. Instead … it flows narrow and deep, in hundreds of steps; revealing greater insights into how social networks serve as a communication medium. This has implications on how social networks are evaluated by intelligence analysts.”

IntelDump quotes from the study abstract:

The dissemination of information is a ubiquitous process in human social networks. It plays a fundamental role in settings that include the spread of technological innovations, word-of-mouth effects in marketing, the spread of news and opinion, collective problem-solving, and sampling methods for hidden populations. The basic models for studying such phenomena posit that information will diffuse from person to person in the style of an epidemic, expanding widely in a short number of steps according to “small-world” principles. However, despite recent studies in online domains, it has been difficult to obtain detailed traces of the dissemination of a single piece of news or information on a global scale to assess the predictions of these models. As such, it has remained an open question whether the spreading of information truly proceeds with a rapid, epidemic-style fan-out or whether it follows a potentially more complex structure. The difference between these possibilities has consequences not only for the models that are used to capture their essential properties but also potentially for the “life cycle” of a piece of information as it spreads through the global social network.

Here, we trace these types of large-scale information-spreading processes at a person-by-person level using methods to reconstruct the propagation of massively circulated Internet chain letters, and from these observations we propose a new set of principles for how such processes work. We focus in particular on two such chain letters, which exhibit tree-like patterns of dissemination that are quite similar to each other but are initially in conflict with the intuitive picture of how information spreads in these settings. Rather than expanding to many individuals in a few steps, the trees are very narrow and continue reaching people several hundred levels deep. We describe a mathematical model that produces trees with this characteristic structure, grounded fundamentally in the observations that social networks are highly clustered and that information can take widely varying amounts of time to traverse different edges in the network. The simple structure of the model, and the fact that it is based on earlier empirical studies of human response times, thus suggests a possible basis for this narrow and deeply reaching style of information transmission in the local dynamics of communication within highly clustered social networks.