Facebook Research’s Artificial Intelligence (AI) researchers have announced the launch of PyText, an open source project dedicated to modeling Neuro-linguistic Programming (NLP), a tool that currently generates more than 1 billion predictions for social network users and Facebook App portfolios.
Specifically, PyText is the technology that supports the voice commands of the new intelligent portal screens and the “M” intelligent assistant of the Facebook Messenger messaging app. For example, the wizard recognizes words that are used in Facebook Messenger chats to suggest actions such as booking trips, a friend’s birthday wishes or recommending recipes.
In addition, Facebook Research announced a number of features for the intelligent portal displays, including a custom camera control that AI uses to move the device to normal video or voice calls. In a blog post by Research, Ahmed Aly Hegazy and Christopher Dewan, researcher and project leader, explained:
“We plan to use PyText as our main NLP platform in the future (…) AI researchers and engineers can now use PyText to test and deploy systems for document classification, sequence marking, semantic analysis, multi-task modeling, and other tasks faster and easier.”
The use of PyText has improved the accuracy of AI communication models in Facebook applications by up to 10 percent. With multiple server and GPU training sessions, PyText has been able to triple the fixed training time.
PyText is embedded in DeepText, another conversation service based on the company’s AI that analyzes all words used by users on the social network. The architecture is based on components that simplify the workflow and enable faster experimentation. The PyText project is now available on GitHub.