Home > Social Media > Blab wants to know what the Internet will love…

Blab wants to know what the Internet will love before the Internet does

Geico - Mutombo Cluster

What if one day, social analytics tools were so powerful they could predict what was going to happen next? Well a new algorithm-powered application called Blab wants to do exactly that by forecasting what is about to go viral.

The last year has seen a slew of social analytics startups hit the scene. Not only do these apps tell you who’s tweeting, retweeting, identify influences, and other rather two dimensional metrics, social media analytics 2.0 like Salorix and Demographics Pro throw machine learning, predictive analysis, and other goodies into an algorithm so they can dig far deeper into user likes, dislikes, habits, and other rather personal details.

And at the far, extreme end of the social analytic spectrum you might remember Raytheon’s RIOT, which uses predictive algorithms to predict your next move before you even make it. Blab doesn’t venture quite into that extreme, into the realm of “real-world” habits, but it is trying to know what the Web will do before the Web does – which can be an extremely valuable piece of information. 

Sourcing Blab’s data

Blab articles

From 50,000 social media and news outlets, Blab has a good idea about what news and viral events break. It pulls data from mainstream social media sources like Facebook, Tumblr, Delicious, Flickr, Twitter, YouTube, and others courtesy of data providers like Gnip, which provides Blab access to Twitter’s Decahose and Tumblr’s Fullhose, Blab’s Chief Technology Officer David Snelling tells me. It also looks at “blog sources, both news related and social.” The list continues and probably won’t cease for some time since the team is continually adding more sources.

But this is just the groundwork.

The engine

Blab category

Blab’s engine is what you’d call a machine learning algorithm. It gets better over time as more and more data is fed into it; it’s also able to learn from its mistakes.

From the beginning of the conversation (the moment it appears), Blab tracks the growth and decline of the conversation by looking at the metadata and the actual text and context. This is then corroborated by its database of 50,000 sources and hundreds of millions of conversations to determine its virality. Blab only needs 30 minutes of conversations and signals to figure out what’s going to happen with a certain event. At the same time, Blab is looking back in its archives (or you can call it a history) to figure out if this event had happened before to figure out how it unfolded. So while we can use common sense and simple tools like Google Trends to watch the Internet life of Grumpy Cat, Blab can pinpoint the moment the meme came into our collective conscious, when it peaked, what conversation or online post captured its momentum, and a host of other information. 

Where it gets interesting is what Blab’s algorithm does with all this data. All of these hundreds of millions of conversations are analyzed and its machine learning algorithm keeps a note about the behaviors that Blab has learned about historical events. And the data is then later put to use to make rather accurate predictions – so maybe Blab can find the next Grumpy Cat before the rest of the Internet does.

The fun stuff: Making predictions

It should go without saying that Blab isn’t omniscient; it can’t predict an unforeseeable event like natural disasters. What it can do, however, is analyze the behavior of a new conversation, and look at previous but relevant conversations to predict whether there’s a chance it could evolve into a viral event.

“The historical data for a conversation is stored as a living entity (chromosome) in our gene pool,” Snelling explains. “It lets us spot and later label certain types of conversations as going viral, being just a blip, or heading down. This is based on the fact that we do associative modeling that lets us understand the probabilities of similar things.”

For instance if you remember Hurricane Sandy, the effect of the disaster going viral meant that brands were jumping on the bandwagon and creating content around the situation (for better or worse). Interjecting a brand into a conversation that everyone is talking about is hard to resist and quite profitable. Blab was testing out its chops during this event and told us that its algorithm “picked up conversations within minutes of first sight and in every language.”

And to make a point of how predictive Blab can be, Snelling adds that while all predictions are forecasted 24, 48, or 72 hours in advance, over time as Blab soaks up more data and learns from previous signals its algorithm will be able to predict cyclical or seasonal events. It could very well be that an event could be predicted months before it happens.

The platform shows users the probability score that updates every few seconds, and predictions are “highlighted” in an orange when they’re above 70 percent probability. Within this Snelling says, “We generally have between an 85 percent and 90 percent or more probability.”

Blab is a marketer’s gold-mine

Another intriguing use for Blab that Snelling points out is that Blab users can purchase Adword keywords that are predicted to trend meaning that you’d be buying these keywords at a bargain. He adds, “there are also API uses for plugging into Realtime Bidding Engines giving a couple of key pieces of information: What words will trend and exactly when will they spike so I know when to spend bigger and when to spend smaller and on what.”

Tools like Blab are clearly aimed at the marketing and PR professions – but given the profitability and immersive nature of the Internet, that’s changing. Everyone is on top of meme and GIF culture, regardless if  it’s for work or personal purposes, and being on top of “what’s next” can be a badge of honor or your next paycheck. Right now, it might be marketers who are buying Blab, but it won’t be long before it’s something anyone with a Tumblr wants to get their hands on. 

Get our Top Stories delivered to your inbox: