Facebook has just announced that its latest artificial intelligence, Deep Text, is meant to understand text as it is inputted into its network. This has many practical applications but, at its most basic, the AI is supposed to learn the meaning and sentiment behind every post.
This latest update on Facebook AI is meant to make the presentation of information more meaningful. For instance, Facebook will now be able to sift through content that the user is actually searching for in the network. Things like spam could be a thing of the past. It might sound minor but this latest AI has the potential to become a very powerful search engine, Quartz reported.
Before Deep Text, Facebook was not able to figure out the intent behind a post. Although Facebook already uses demographic information through direct and indirect interactions with brands, the information gathered did not have structure. Deep Text is going to change all that.
Hussein Messana, director of Facebook's Applied Machine Learning team, said to Quartz, “We want Deep Text to be used in categorizing content within Facebook to facilitate searching for it and also surfacing the right content to users.” He added, “If we can understand text, we can help people connect and share in a lot of different ways.”
One of the applications will be to understand when a user is posting something about selling a product. Details like price and others can then be pulled to make the post better geared for this purpose.
Deep Text will be able to track all the information poured into its servers. The same principle is used by Google now to understand the questions being asked of it. This will then give users the right answers to their questions.
Searching Facebook for “laptop,” for instance, already yields posts by friends on “laptop” and even provides laptop-related posts from brands. Deep Text could possibly also now provide a recommendation where to buy a new laptop based on posts made by friends.
Deep Text is based on neural networks. It can sift through and understand thousands of posts per second. It can also understand 20 languages “with near-human accuracy,” as Tech Crunch reported. Deep Text is also able to learn semantics between words. The example used were “brother” and “bro” which are typically used in similar situations. The millions of pages created by users will also be used to teach Deep Text.