Today in Technology
Earlier this week, Facebook introduced DeepText. DeepText is Facebook’s new text understanding engine which was developed in order to surface new user created content, such as status updates, and filter spam. It is an approach that processes natural language using the deep learning framework. The new engine is a learning-based text engine which understands, with human level accuracy, the textual content of thousands of posts per second. DeepText was developed using several deep neural network architectures which can perform word-level and character-level based learning.
By utilizing the deep learning architecture DeepText is able to understand variations within each language such as different spelling or use of slang to communicate an idea. The use of Facebook’s deep learning architecture allows DeepText to learn from text with little pre-processing effort. DeepText is able to understand the meaning of “I want a blackberry” by using context clues – it can tell if the user is posting about the fruit or the phone.
In a traditional NLP approach, words are converted to numbers which are read by a computer algorithm. There is a huge downside to this approach: the spelling of words – slang words are not associated with the ‘original’ word because it is spelled differently. Facebook’s DeepText uses a mathematical concept which associates slang words with their original meaning. This allows DeepText to understand semantics across multiple languages. Eventually, the deep learning architecture refines these word groupings into smaller labeled data sets.
DeepText is useful for users because it helps users express their thoughts and classify their posts in an orderly manner. This feature shows users posts relevant to their needs which indirectly increases post engagement – you are likely to share, like, or comment on a post that is relevant to your interests. DeepText is also advantageous to Facebook’s ad-targeting. With the use of DeepText, Facebook can tailor ads directly to the users wants and needs.
Is DeepText valuable to Facebook users or is it another way for Facebook to archive information about users? Comment below voicing your opinion on Facebook’s DeepText.