On Facebook, we routinely fire off dozens of posts and comments per week, with the assumption that the recipient will be able to decipher our messages. But what about Facebook's algorithm? Is it powerful enough to decipher your text-based chatter and help improve your experience on the social network?
That's where Facebook's new DeepText AI comes in. It comprehends "with near-human accuracy" the textual content of several thousand posts per second, in more than 20 languages, to help you call an Uber, sell used furniture, or ensure you never miss your favorite celebrity's posts.
Led by Facebook engineers Ahmad Abdulkader, Aparna Lakshmiratan, and Joy Zhang, the project aims to help improve users' experiences, "whether we're surfacing more of the content that people want to see or filtering out undesirable content like spam," they wrote in a blog post.
Traditional neuro-linguistic programming (NLP) converts words into numbers a computer algorithm can read. Deep learning, however, uses "word embeddings" to preserve the semantic relationship among lexicon. It also applies across multiple languages.
DeepText has already been put to the test in some Facebook functions: Messenger, for example, is getting better at understanding when someone needs to go somewhere. It can recognize the difference between "I just came out of the taxi" and "I need a ride."
Facebook is also using high-accuracy, multi-language DeepText models to help users peddle items by extracting meaningful information such as the object being sold and its price.
Write "I would like to sell my old bike for $200. Anyone interested?" on the social network, and it will automatically help you create a sale post—an attempt to stand out against the throngs of conversation filling users' feeds.
The same goes for finding and surfacing the most relevant or high-quality comments (in multiple languages) among exchanges with celebrities and public figures.
"We are continuing to advance DeepText technology and its applications in collaboration with the Facebook AI Research group," the company said.
The privacy implications of this service—which examines posts, photos, and videos for text—remain unclear. Facebook just last month was targeted by a class-action lawsuit alleging the tech titan violated federal privacy laws by scanning private messages.
DeepText is based on ideas originally developed in papers by Ronan Collobert and Yann LeCun from Facebook AI Research.