Facebook Taps Artificial Intelligence to Enhance Text-Based Search Features

In a world where an enormous and continuously growing portion of the population uses Facebook, the social media giant seeks to effectively handle the vast amount of information it receives, with around 4.1 million likes noted for various posts every minute of the day. The company aims to develop tools that help users locate posts that align closest to their interests amidst its colossal ocean of data. This led to the development of the News Feed algorithm that prioritizes posts from friends above page posts. Taking this a level further, Facebook recently announced the advent of Deep Text.

Deep Text is an avant-garde Artificial Intelligence (AI) tool designed to comprehend the semantics and sentiments embedded in user posts. Its introduction implies that users may soon find their feeds filled with posts that directly correlate with their specific interests, determined by their past activities on Facebook. This was published in a blog post on Wednesday. Facebook defines this groundbreaking AI system as a “deep learning-based text understanding engine”, one that can interpret the textual content of several thousands of posts per second with near-human accuracy across more than 20 languages.

This AI tool has massive potential to revolutionize the Facebook search experience and eradicate spam messages. Facebook houses trillions of bytes of data across posts, photographs and comments. Deep Text may be able to sieve through this expansive database to provide users with precise information relevant to their interests through intelligent learning.

“In conventional Natural Language Processing (NLP) methodologies, words are translated into a format comprehensible for a computer algorithm. For example, the term “brother” might bear the integer ID 4598, whereas “bro” may carry a different integer ID, such as 986665. This stipulates that each term must feature in the training data with exact spellings to be understood,” explains Facebook in the blog post.

“However, with deep learning, we can leverage “word embeddings,” a mathematical tool that retains the semantic relationship among words. For instance, we can calculate that the word embeddings of “brother” and “bro” are spatially close, enabling us to capture the deeper, semantic definitions of words. Using these word embeddings, we can comprehend similar semantics across diverse languages, even with surface form differences. For instance, in English and Spanish, “happy birthday” and “feliz cumpleaños” are near each other in the common embedding space.”

The utilization of Deep Text stretches beyond improved search facilities by directing users towards specific Facebook tools that assist them in accomplishing their tasks. For example, should a user mention upcoming movie plans for the weekend, Deep Text can swiftly comprehend this and guide you towards relevant apps or posts.

In 2013, Facebook introduced Graph Search, developed in collaboration with Microsoft’s Bing, which was later discontinued in 2014. It functioned based on algorithms that accessed its Big Data to provide efficient search results centered around Facebook. This marked Facebook’s preliminary step towards becoming an enhanced search engine on its own terms and striding away from functioning purely as a social networking platform. However, underpinning it with Deep Text could potentially multiply the data Facebook offers its users, spanning from photos, videos and articles, further enhancing the user search experience.