Facebook black science and technology will be able to analyze thousands of articles a second

The goal of

Facebook is to understand the user. In the Facebook social network, every day there will be a lot of data to be shared, Instagram, WhatsApp or even Oculus, for Facebook, the urgent need to find a way to catalogue all the information, and the information is easy to search for the final, but obviously this is a very difficult task. Therefore, Facebook will focus on artificial intelligence.


Based on this

, Facebook days ago released a new product DeepText of artificial intelligence, it can accurately identify human chats, DeepText from Facebook AI Research group, using a variety of depth of neural networks, can make text and character level collocation learning, deep learning is used to train the model, can be used to understand Facebook Messenger user dialogue or Facebook user posts, provide advice to understand user needs and interests.

Facebook pointed out that compared with the traditional Natural Language Processing technology, deep learning oriented technology enables DeepText to rely on less knowledge of language, text meaning can have a better understanding of different languages, and effectively deal with the label of data, to meet the Facebook in the text processing on demand. In addition, Facebook also said that DeepText can be close to the accuracy of human understanding of people’s chat content.

in addition, relying on the computing power of the background, this tool can be analyzed every minute to identify thousands of users comment or chat. In terms of language, this tool has been able to analyze more than 20 languages. In addition to identifying chat content, given the automatic recommendation, DeepText is another area of application to combat spam messages. In the social network platform, there are a lot of network marketing posts, Facebook can automatically identify and quickly delete.

DeepText will be integrated in the mobile chat tool, the system can automatically analyze the user’s chat information, such as if you think the two friends are talking about a travel plan, the system can automatically match the Uber, Lyft and other mobile car information, or give the weather forecast etc..

on the next step in the future development, DeepText R & D team will allow DeepText: to better understand the user’s interest, so as to provide users with their interests more closely match the content. Having the ability to combine text and image content. The development of new depth neural networks such as bidirectional recurrent neural network (BRNN).

new wisdom yuan in a recent article, summed up:

future, if DeepText >