Twitter is now using a trendy type of AI to figure out which tweets to show you
Twitter’s engineers say ‘deep learning’ increases tweet engagement and time spent on the platform
Twitter today said it has begun to use artificial intelligence (A.I.) to recommend certain tweets in users’ timelines.
The announcement comes a week after billionaire Mark Cuban told CNBC that he had recently begun buying Twitter stock because he believed “they finally got their act together with artificial intelligence.”
Facebook, Google, Microsoft, and other companies have previously attempted to improve various products using deep learning, a trendy type of A.I. Twitter has brought on people who are talented in this area through acquisitions of companies such as Magic Pony, and it has open-sourced some of its deep learning software. But the company has not been especially transparent about its progress.
A year ago Twitter introduced a so-called algorithmic timeline that ranked tweets based on relevance instead of them being in reverse chronological order. The feature is on by default, and users can opt out to revert to the classic timeline style. But Twitter has also introduced various widgets near the top of users’ reverse-chronological timelines to show off tweets that it thinks users might like or might have missed.
Before putting the deep learning system into production recently, Twitter was using less computationally intensive machine learning methods such as decision trees and logistical regression, Twitter software engineers Nicolas Koumchatzky and Anton Andryeyev wrote in a blog post.
For its 328 million monthly active users, the company is evaluating and scoring thousands of tweets per second to determine what’s worth recommending in timelines, taking into consideration an increasing number of factors, including whether tweets contain images or videos, the number of retweets and likes, and your previous interactions with other account holders, Koumchatzky and Andryeyev wrote.
“Online experiments have also shown significant increases in metrics such as Tweet engagement, and time spent on the platform,” the engineers wrote.
For the past two years Twitter stock has stayed below US$40 per share after closing at US$44.90 on its first day of trading in 2013. In the past year some members of Twitter’s Cortex A.I. research group have left; in March the group’s technology lead, Clément Farabet, took a job at Nvidia.