1 Comment

Summary:

Who needs a Ph.D. in statistics when you have the cloud? Machine learning is high data science, and it’s fast becoming something that anyone leverage to sell more handbags, or solve a research problem, or build the next LinkedIn or Facebook.

Who needs a Ph.D. in statistics when you have the cloud? Machine learning is high data science — a discipline focused on algorithms that automatically detect complex patterns hidden within datasets — and it’s fast becoming something that anyone leverages to sell more handbags, or solve a research problem, or build the next LinkedIn or Facebook.

Machine learning is nothing new — companies have been using it for years for everything from market predictions to search engines — but its presence in the mainstream is. Earlier this year, we covered software vendors such as HPCC Systems and Skytree that are trying to make machine learning easy for all types of companies. Software, however, still requires hardware, licenses and all the things that go along with it. And maybe even some analytics expertise.

If you’re a small business or maybe even an individual, you need something easier. You need something in the cloud. And you need someone else to handle the hard parts.

Marketers, we have you covered

These somebodies are already popping up for specific use cases, especially web marketing. We’ve covered some of them already, including BloomReach, InsightsOne and even Bookt. As it turns out, getting consumers to buy things (well, your things) online isn’t always easy. Machine learning can help by finding those hidden patterns you probably didn’t even know existed, so you can tune your marketing campaigns accordingly. Maybe even make them personal.

One startup trying to do just that is DataPop, which just closed a $7 million Series B round for its service that tries to present the most-relevant search ads to each individual consumer. DataPop relies on semantic search and natural-language processing to infer connections between what consumers enter into the search window and what they really want, and then on machine learning to help with everything from determining common spelling mistakes to search construction to the sequence of events that leads to a purchase.

“When we can understand the structure of these [ad] campaigns,” DataPop Co-Founder and COO John Zimmerman told me,  “… that provides us with the data to actually do the math and understand what’s happening where.”

That type of insight can be invaluable to a small, niche company selling designer handbags, for example, but the company doesn’t have to do anything but feed its inventory data to DataPop. Then DataPop’s algorithms will handle the task of figuring out the best message — price, products, brand names, etc. — to place in front of each consumer based on the customer’s search terms. DataPop will run tens of thousands to hundreds of thousands of unique ads per campaign, Zimmerman said, and that’s “just not possible for a human to do.”

A human could never understand all the patterns and relationships at play in the data, in part because they’re so subtle. “It’s not this big beat-you-over-the-head-with-an-overaggressive-message [approach],” Zimmerman said. “There are certain triggers that will get people to respond.”

One man, one spreadsheet and a few mouseclicks

But even for relative laypersons not in the marketing business, machine learning isn’t out of reach. Recently, someone alerted me to a new cloud service called BigML, which aims to let anyone do machine learning in four easy steps: set up a data source; create a dataset; create a model; and generate predictions. It’s in private-beta mode right now, but the service has promise because it’s so simple (you click buttons, you don’t write algorithms) and so visual.

Here’s what a sample model for determining the likely fuel efficiency for a car looks like. BigML models are interactive (even the sample ones), so you can investigate the patterns the service detects just by scrolling over and clicking on the nodes.

It’s amazing, quite frankly, how fast the big data space is evolving, especially with cloud services targeting specific use cases and smaller users. While large enterprises toil through complex and massive Hadoop deployments, someone with a CSV file can apply machine learning algorithms to it in a matter of minutes, or a mom-and-pop e-commerce site can hit the right user with the right ad without lifting a finger. This is how big data can make a big difference, from the bottom up.

Feature image courtesy of Shutterstock user Andrea Danti.

  1. Check out our full day Large Scale Machine Learning event in SF with Carnegie Mellon University, Twitter, Netflix, Pandora, IBM Watson, and more. http://ml10.eventbrite.com

    Share

Comments have been disabled for this post