Predicting When Things go Viral

How can we do this? How accurate can it be?

Posted 1 year ago by Tom Maiaroto.

Predictive analytics has to be one of the most interesting and desired things in the tech industry right now. I remember well over a year ago at a meetup, where Kevin Rose was speaking, talking to a guy who couldn't care less about social media analytics. It was a good turn out given the speaker and I figured I'd run into some people who really had some good opinions. So this guy was raving about predicting user behavior. What people will do and how they will react. However, as I was talking to him about my older analytics platform, he simply dismissed it. As if it had no value (of course we know better, given the success of Radian6, etc.). He was just so focused on something he couldn't possibly understand, but so desperately wanted.

I've written about this before a little bit. I firmly believe that when you start getting into crystal ball space, you have to be very careful. People are too quick to rush into this and they don't fully understand what the data is saying or predicting. Like this guy, many others are quick to rush past the metrics of "now" and just want to get to what "what will be."

The simple truth of the matter is if you don't understand the "now" then how can you ever possibly begin to understand what will be? How can you act on it? In order to truly understand the future we must also understand the past.

We have a large enough problem right now with current data. If you've been following along Virality Score, it should be apparent that most people can't even accurately measure what's viral this very second. Furthermore, they don't know when it ceases to be viral nor do they know to what degree it is viral. So now tell me, how would one in this position ever be able to make a prediction?

I purposely started off solving a very real world, complex, problem in a simple and elegant manner. Only now with this data can I begin to even attempt to make predictions. Any prediction on otherwise inaccurate information would be even further off base.

Mashable has been working on their "velocity graph" for a little while now and I'm glad to see some effort put into this area of research. They recently created an Android app claiming to predict when their news goes viral (keep in mind, just their news).

I was excited about this app because their velocity graphs never had any context. You see a squiggly line and of course orange means good, right? We don't know how good, but we can see the line going up so it must be good. Granted, I have no doubt they are recording shares over time, but even that does us little good.

So what did this predictive app show us then? Now they're putting their money where their mouth is. If I have something alerting me to when something is about to become viral...Then I expect it to become viral (to at least some degree). First, I quickly put together a little ranking of their content. This will update every now and then with an actual accurate degree of virality for each story. I also put a cute little sparkline in there of my own for fun.

You could literally use this page to compare against their velocity graphs (I'm taking the last 10 hours in those graphs by the way...Who knows the range for theirs). However, what I really wanted to do was wait for their app to send me a notification of what they felt was going to be viral. It does from time to time. It just told me this amusement park story was going viral. So I run over to my site here and take a look. The degree to which that story is viral is 5.26, or in other words it is more viral than 5.26% of the internet. That's actually not too bad. There's just one problem. It used to be more viral than that. You can see in the screenshot (because the graph will change by the time you read this) that it was even more viral within the past 10 hours. However, you can clearly see that it is trending up again.

So I'd have to say that Mashable's app does a good job of alerting you to stories that are getting some social media love.

Are there stories that are more viral? Yup. Did I get push notifications about all of them? Nope.

So is some of this curated? Do they give certain articles preferential treatment? Maybe. We can't really assume anything. It's a little vague here. All I can say is that their app seems to work, but doesn't do as good a job as I'd like personally (in this department). Their ranking under what's "hot" isn't quite accurate, but it isn't bad either.

However, their app does serve as a really good way to read their content (most important feature) and I think other news outlets would be wise to follow suit with what they are trying to accomplish here. Especially news sites that like to write the same stories over and over. Seriously? As readers, we don't have time to weed through all of the, somewhat, duplicate content.

Their app definitely has a place on my device and I enjoy using it. Maybe even more than their site...But in this case, because I want a clean way to read their stories and not because I turn to it to tell me what's viral.

So the ultimate question here is, can Virality Score predict? Yes. It can. How accurate will it be? We'll just have to wait and see. However, the integrity and accuracy of the ranking data (the viral measure) - that foundation - serves as a very strong springboard for such predictions. Again, skipping steps here can really hurt you. Patience is important.

Filed under
comments powered by Disqus

Search for Posts

Popular Labels

social media virality score general internet web development reviews and opinions php web tools hosting lithium web design

Recent Posts