On Wednesday, Facebook announced their newest tool to better understand their user base: DeepText. Using Deep Learning algorithms, techniques, and neural nets established in several research papers, the technology is designed to learn and understand how humans use text in the context of normal conversation.
As explained on their blog, this is how DeepText works in a nutshell:
“DeepText leverages several deep neural network architectures, including convolutional and recurrent neural nets, and can perform word-level and character-level based learning. We use FbLearner Flow and Torch for model training. Trained models are served with a click of a button through the FBLearner Predictor platform, which provides a scalable and reliable model distribution infrastructure. Facebook engineers can easily build new DeepText models through the self-serve architecture that DeepText provides.”
The most immediate benefit for Facebook is how it may affect advertising for certain users. For example, I recently got engaged and announced it on Facebook, and I’m still seeing ads for Brilliant Earth engagement rings. I know I need to be sold to so Facebook makes money, but maybe they can be a little better about it. Ideally, with DeepText, advertising can be switched around quickly if a few mentions are made about a product they already own. DeepText can also learn about a users’ interest to display more content in their news feed that’s more relevant to them from pages and people they follow.
The project is currently in use with Facebook’s messenger application to recognize when someone might want to go somewhere, and offer taxi or ride sharing ads (you can see a video demonstrating this use case on their announcement post). Another use is to point users in the direction of existing ways to sell items on Facebook if they’re looking to sell something.
There may be a few disconcerting elements to this news, such as the blatant analysis of your private messages (which was probably happening already), but it seems like Facebook is trying to make its site more engaging not just for advertisers, but between users sharing content as well. Just don’t be surprised if it seems to know you a little too well.