Facebook on Wednesday unveiled DeepText, a deep learning-based tool it's using to help it make sense of the mountains of unstructured data that live on the social network.
The text understanding engine, Facebook says, can understand with near-human accuracy the textual content of a thousand posts per second, in more than 20 languages.
"Understanding the various ways text is used on Facebook can help us improve people's experiences with our products, whether we're surfacing more of the content that people want to see or filtering out undesirable content like spam," the company said in a blog post.
DeepText is already being used for some applications, such as Facebook Messenger. For example, if someone types that they need a taxi, the tool can help send the user a prompt to call a taxi. Facebook is also beginning to use DeepText models to point users to various tools based on their posts.
It fits into Facebook's midterm and longer-term plans to use artificial intelligence to enhance its core ecosystem and eventually branch out into new ventures.
In the meantime, the next steps for DeepText include using the tool to monitor comments -- that includes surfacing high-quality comments while removing objectionable ones -- and recommending relevant content to people. Facebook is also building new deep learning architectures that can understand text and visual content together.
The company is also investigating new deep neural network architectures, it said, such as bidirectional recurrent neural nets (BRNNs).