Find out what the top machine learning platforms think about you
We built our gender classifier in CrowdFlower AI, our new beta offering. For this job, AI makes predictions on whether you're a man, woman, or brand using four of the best machine learning platforms out there. They are:
- Scikit: The most popular open-source machine learning tool
- Metamind: A deep learning startup specializing in image recognition
- Google Predict: Google's ML offering
- IBM Watson: IBM's tool that famously beat Ken Jennings on Jeopardy!
Where did this data come from?
For this project, we used Twitter's API to pull 20,000 random tweets along with the user's profile, their description of themselves, and a few other attributes. We then showed those 20,000 profiles to our contributors, who determined whether the account belonged to a man, a woman, or a brand. We used those human judgments to train our model.
Where do these predictions come from?
CrowdFlower AI looks for patterns in datasets. In the use case above, certain words, emoticons, and even link colors inform the forecasts it makes about a profile. You can read about some of the common predictors in the blog post we authored about the project, or download the original dataset.
What do the percentages mean?
The percentages in the widget above are confidence scores. The higher the number, the more confident our model is about its accuracy.