Facebook AI Digs Deep Into User Content
Jun 3, 2016 9:12 AM PT
Facebook on Wednesday introduced DeepText, an artificial intelligence-fueled text analytics engine.
"Text is a prevalent form of communication on Facebook," wrote Facebook software engineers Ahmad Abdulkader, Aparna Lakshmiratan and Joy Zhang in a post describing its capabilities.
"Understanding the various ways text is used on Facebook can help us improve people's experiences with our products," they continued, "whether we're surfacing more of the content that people want to see or filtering out undesirable content like spam."
DeepText can understand with near-human accuracy the textual content of several thousands posts per second, spanning more than 20 languages.
The engine leverages several deep neural network architectures and can perform word-level and character-level based learning, noted Abdulkader, Lakshmiratan and Zhang.
DeepText's neural networking techniques have advantages over traditional natural language programming approaches, which are very literal. Deep learning can puzzle out the deeper semantic meaning of words in multiple languages.
"With traditional methods, you have to preprocess the data before it's usable. With text, that means adding language-dependent features like grammar, vocabulary and parts of speech," explained Eli David, CTO of Deep Instinct.
"When you do the same thing with deep learning, you just feed in a lot of text from a language, and the model learns by itself. It's similar to the way a child learns language," he told the E-Commerce Times. "The big advantage of DeepText is that it can easily be expanded over a number of different languages which would not be possible using traditional methods."
Moreover, deep learning can produce accuracy rates that are 10 percent to 20 percent better than traditional approaches, David pointed out.
Better Than Off-the-Rack
While Facebook has tons of content, it hasn't always been able to use it to its best advantage.
"This effort should allow it to make use of up to 100 percent of the content users post in a way that will allow Facebook to understand what its users are interested in, and serve up more of the same," said Jan Dawson, chief analyst at Jackdaw Research.
"That, in turn, should help drive engagement and time spent, which in turn drives ad revenues," he told the E-Commerce Times.
"It's a technological equivalent of having your clothing custom-fitted rather than grabbing a suit off a rack," suggested Charles King, principal analyst at Pund-IT.
"If DeepText is successful, users will find their Facebook experiences becoming increasingly enjoyable and satisfying," he told the E-Commerce Times. "That won't happen overnight -- it's more a matter of changes that occur over the course of months and even years."
DeepText will allow Facebook to deliver a more personalized experience to its users, said Elizabeth Lampert, president of Elizabeth Lampert PR.
"It knows from analyzing your posts what you're likely to be most interested in and feeds it to you," she told the E-Commerce Times. "Through this, Facebook definitely secures its position as a leader in the AI field."
It also keeps the social network in the race for acquiring eyeballs through search, "which is huge," Lampert said. "We keep hearing about bots and AI. Both Facebook and Google want to build better AI-based assistants. Facebook's DeepText is a big step in that direction and the one to watch."
What Price Privacy?
User satisfaction isn't the only benefit DeepText could deliver to Facebook.
"Facebook is also likely to use the results of DeepText to further refine advertising and other services it provides to its commercial customers," said Pund-IT's King.
"DeepText will certainly provide Facebook sophisticated tools for mining its information assets, but DeepText will also allow Facebook to more effectively mine its users and monetize their Facebook experience," he pointed out.
"It's unlikely that Facebook would be investing in DeepText without plans to monetize it in numerous ways," King suggested.
"DeepText can have significant implications for ad targeting," said Lampert. "In fact, it's an advertisers' dream come true."
With greater customization, however, come questions of privacy.
"There's definitely a creepy line with most of these things, and Facebook will have to be careful how it talks to users about this technology," said Jackdaw's Dawson.
Still, "given that this is all content users are willingly sharing on Facebook to begin with, it's hard to argue that it's invasive," he added.
"Of course, the privacy implications might rattle some people," Lampert acknowledged, "but giving up some data is the price of doing business in the social media world."