The Maths of Memes, revisited.

Research on using Medical Virus maths for email memes, from the New Scientist:

The viral spread of information online has conventionally been modelled using epidemiological tools developed to analyse the spread of biological viruses. One of the concepts borrowed is that of an infection's Ro, or basic reproductive number, which describes how many other people someone with the virus can be expected to infect.

Knowing the Ro number help predict the likelihood and extent of real life epidemics, such as H1N1 swine flu. But models that apply the idea to online information can only indicate whether an internet meme is likely to be successful or to die out quickly, says Esteban Moro at the Carlos III University of Madrid, Spain.

Moro, working with José Luis Iribarren at IBM in Madrid, used IBM's company email newsletter to show the importance of variations between people's infectiousness in propagating memes online.

They started a reward scheme offering prize draw tickets for recommending the newsletter by providing email addresses of other people and tracked how widely and quickly the recommendations spread. After two months it had reached 31,000 people.

But while people took 1.5 days to respond to a recommendation email on average, there was a huge variation at the individual level: some users responded within minutes, other in months, says Moro.

And only by combining some expectation of that variation with the Ro number is it possible to build a model able to predict the meme's spread. The team use a small chunk of the initial data on the content's spread to predict how many people it will reach in total, and how fast. "Our model can give predictions within 1 per cent error once secondary reproductive number and human activity are estimated," Moro says.

But, you have to measure from early performance in the wild:

The model cannot predict whether a piece of content will go viral before it has been released; only its likely reach once it starts spreading. And the researchers think their approach to modelling should apply to information spreading via social networking sites and other online services as well as email.

The really interesting question, of course, is to try to break down the "DNA" of a viral meme and work out which bits really can infect effectively. But this area of study is called memetics and has been around for at least 20 years (I've been on the discussion groups at least 10 years....)

So apologies if If I may be a bit disrespectful - but there is a certain element of old wine, new bottles here - Memetic Algorithm wonks have been trawling this field for at least a decade - just try Wikipedia for a starter class, and there have been discussion groups, learned journals and conferences on the subject for at least 10 years. And you'd never believe it, but one of the first things they looked at were medical virus models. I don't know if this is just one branch of science rediscovering what another has known for ages, or just confirming work over bigger internet than before (unlikely as its based on representative samples), but it seems - yet again - a fascinating example of how, even in our connected age, great chasms and barriers occur in knowledge transfer from one arena to another

Update - nice pointer from Adriana Lukas, looking at diffusion of memes in Facebook (See video of lecture here), synopsis is:

Using a dataset of 262,985 Facebook Pages and their associated fans, this paper provides an empirical investigation of diffusion through a large social media network. Although Facebook diffusion chains are often extremely long (chains of up to 82 levels have been observed), they are not usually the result of a single chain-reaction event. Rather, these diffusion chains are typically started by a substantial number of users. Large clusters emerge when hundreds or even thousands of short diffusion chains merge together.

This paper presents an analysis of these diffusion chains using zero-inflated negative binomial regressions. We show that after controlling for distribution effects, there is no meaningful evidence that a start node’s maximum diffusion chain length can be predicted with the user’s demographics or Facebook usage characteristics (including the user’s number of Facebook friends)

.