Facebook drives into neural-network-based AI with DeepText

Comprehension engine can understand the context of posts, claims Facebook

Social networking giant Facebook has released a neural-network-based artificial intelligence tool called DeepText that, it claims, can understand the context of people's online posts in more than 20 languages.

The backbone of the software is the FBLearner Flow and the Torch open-source machine learning library to perform word-level and character-based learning.

According to Facebook, the system can understand slang and make sense of potentially ambiguous phrases. For instance, if a Facebook user posts the phrase ‘I like apple' DeepText can work out whether it refers to the fruit or the company.

Facebook had to go beyond normal neuro-linguistic programming (NLP) techniques with DeepText, it says, as the extensive pre-processing logic built on top of intricate software engineering and language knowledge is ineffective at picking up variations in languages and spelling when people post on the same topic.

Neural-network deep learning enables DeepText to learn as it sifts through Facebook posts, bypassing the need to rely on pre-programmed language-dependant knowledge.

"In traditional NLP approaches, words are converted into a format that a computer algorithm can learn. The word 'brother' might be assigned an integer ID such as 4598, while the word 'bro' becomes another integer, like 986665. This representation requires each word to be seen with exact spellings in the training data to be understood," claimed Facebook.

"With deep learning, we can instead use 'word embeddings', a mathematical concept that preserves the semantic relationship among words. So, when calculated properly we can see that the word embeddings of 'brother' and 'bro' are close in space. This type of representation allows us to capture the deeper semantic meaning of words."

DeepText is being tested across Facebook's various services, including Messenger, which is helping it to become smarter, such as understanding when someone might be expressing a desire to go somewhere.

Facebook sees potential for DeepText to improve the experiences of Facebook users, for example by automatically detecting and removing spam posts or displaying the most relevant comments.

Facebook's work on deep learning technologies is pertinent given that the company is developing an AI-powered virtual assistant called M for Messenger that mixes advanced machine learning with human customer service.