Saturday 9 July 2011

Big seed marketing

Social media marketing departments must decide how to allocate their budget between making content that will spread - and distributing that content.

The initial distribution is sometimes called "seeding", and focusing on that distribution is sometimes called using a "big seed".

Big seeding was popularised in 2007 by an article entitled Viral Marketing for the Real World - by Duncan J. Watts, Jonah Peretti, and Michael Frumin.

So, in the light of their article, the question naturally arises: does big seeding actually work?

For my epidemic threshold article I made a simple computer simulation of epidemics that produced graphs showing how the number of infected individuals could increase or decrease over time.


An epidemic - showing the number on infected hosts plotted against time.

The model that produced the above diagram is extremely simple. Individuals are modelled as being either infected or not infected. They have a constant probability of dying in each generation. Senescence and pathology are not modelled. Infected agents infect more agents (randomly) in each generation. Then some individuals (chosen randomly) die, and are replaced by healthy newly-born individuals. The population size is a fixed constant - so the death rate and birth rate are equal. The plot was made by varying two parameters: the infection rate and the death/birth rate. These variables are sampled from a uniform bounded random distribution to create the plot.

The seed population is fixed at the same value for each run and is shown as the y-intercept on the left hand side of the diagram.

The diagram illustrates the concept of an epidemic threshold - if content is insufficiently infectious, it dies off, and goes extinct.

The next issue I wanted to explore was to see how the seed population size influenced the extinction rate.


This is a plot of survival against seed population size.

Here, "survival" refers to having a population size of at least 1 at the end of the run - i.e. it refers to not going extinct.

This graph illustrates two main things:

  • Having a seed population too small is often fatal - random fluctuations in population size too easily cause your seed population to execute a random walk into extinction.
  • Big seeding rapidly runs into diminishing returns - provided you seed on a reasonable scale, success depends quite a bit on how much you exceed the epidemic threshold by - and not so much on the size of your seed population.

How to manage the tradeoff between the seeding budget and the contagiousness budget is beyond the scope of this article - but hopefully these graphs will help people to understand the basics of the dynamics involved.

References

No comments:

Post a Comment