How to Model Viral Growth: The Hybrid Model | Hacker News Reader

How to Model Viral Growth: The Hybrid Model

linkedin.com

57 pointsnikunjk7 comments

I actually have a paper published on this subject [1], except we use models that have a basis in epidemiology. That article is unfortunately behind a paywall, but here's an unbranded copy of an early draft of the paper [2].

We base our work off of the classical SIR model [3], which we extend to include the possibility of reinfection (since it's possible to become reinfected with a meme after initially losing interest). The key features of the SIR model is a rate of infection proportional to the social interaction between infected and susceptible individuals, where infected individuals gradually lose interest (or not, depending on how the parameters are set).

With appropriate values for parameters, our model can very closely fit the characteristic curves from viral infections, particularly those with an initial large spike followed by a gradual taper, as you can see from these Google Trends data [4]. The model can also be used to predict future infection levels (assuming no change in system dynamics).

[1] http://www.sciencedirect.com/science/article/pii/S0307904X11...

[2] http://stash.synchroverge.com/files/viral_memetic_model.pdf

[3] http://en.wikipedia.org/wiki/Compartmental_models_in_epidemi...

[4] http://www.google.com/trends/explore#q=two%20girls%20one%20c...

This seems trivial since it's assumed the users are "immortal" - never leaving the product.

Maybe I'm not the target audience for this article though.

I plan to build up to a relatively complex model that hopefully you'll find useful or interesting.

Here's my intended sequence: for the next post, I'll model retention as a simple loss rate. After that I'll model retention with a curve over time. And after that I'll model the viral factor itself as a curve over time.

Is there something in particular you think it'd be interesting to cover?

It would be great if you could consider the cycle time.

As someone with a lot of experience modeling growth, I think this teaching approach is a good one.

I find that most teams start exactly where this article starts - modeling total installs over time using constant virality. They then realize they care more about active users, and do the same for retention, then realize that retention and virality change over time, and do a cohort model.

Jumping to the end is likely to confuse people who haven't already organically gone through this process. I was also a little underwhelmed at first read, so some mention of the plan in future posts would help us know where you are going.

"As simple as possible, but no more"

--Some famous guy. =]

And here I was hoping for a tutorial on virology viral growth curves ;o(

Reply on news.ycombinator.com