AI

By Justin Tenuto, November 4, 2015

Announcing CrowdFlower AI

A few weeks ago, we made an announcement at the Rich Data Summit. It deals not just with where we want our company to go, but where we think machine learning and data science are going in the coming years. Which is to say: human-in-the-loop machine learning. 

What we’re working on is something called CrowdFlower AI. We’ll be talking about it a lot in the coming months and we’re really thrilled with what it will allow our customers to do. Essentially, you’ll be able to run CrowdFlower jobs like you always have, but then take the enriched data our platform and contributors provide and use it to train machine-learning algorithms. One of the most powerful parts of this is that the algorithm CrowdFlower AI will create will be based on your specific use-case, which gives you much, much higher accuracy. Here’s how:

Take the term “addictive,” for example. If you were running a sentiment analysis job about, say, a new video game you released, that term is definitely positive. It means people can’t put your game down. But if you were talking about cigarettes, that same word carries a vastly different, vastly more negative connotation. That’s just one reason why using out-of-the-box natural language processor gives you lower accuracy than training an algorithm on a specific use-case. Which is to say: context matters. A lot. 

Of course, it’s not only for sentiment analysis. CrowdFlower AI uses text to create models, so you can do anything from categorizing help tickets to pruning your sales leads.

Want to see how it works? We’ve got a video of the talk where we announced it. Check it out:

If you’d like to sign up for our beta or to get additional information, just head to our sign-up page. We’ll be showing everyone how CrowdFlower AI works in the coming months, as well as highlighting some of our favorite talks from the Rich Data Summit (where, serendipitously, a lot of folks were talking about human-in-the-loop machine learning). Stay tuned and, as always, thanks for reading.