Have you ever wondered how Twitter ranks real-time updates, missed tweets and recommendations in your timeline?
Look no further, as the top microblogging site has finally revealed how its tweet ranking algorithm works.
Nicolas Koumchatzky, staff software engineer at Twitter’s Cortex team, outlines how the company continues to improve tweet ranking over the past year. He is part of the AI team that uses machine learning to leverage the platform’s artificial intelligence.
Tweet Ranking in Timelines
Before the algorithm, Twitter says your timeline included all tweets from people you follow in reverse-chronological order since your last visit.
Then tweet ranking came and changed how tweets enter your streams. It predicts how interesting and engaging tweets are for you based on scores. And the highest-scoring tweets will eventually land at the top.
In addition, Twitter shows a dedicated “In case you missed it” module based on the time of your last visit and the number of tweets available. The module contains few of the most relevant tweets based on their scores.
“The intent is to let you see the best Tweets at a glance first before delving into the lengthier time-ordered sections,” writes Koumchatzky.
Criteria for Judging
Twitter considers the following factors when scoring tweets:
- Timestamp of the tweet, presence of images or videos, and total number of likes or retweets
- Historical relationship and interactions with the tweet’s author
- Tweets you liked or retweeted and how often you used Twitter
Each time you open or refresh the timeline, Twitter quickly scores each tweet from people you follow since your last visit.
Koumchatzky adds that this list will continue to grow, depending on future interactions, user behaviors and features on the platform.
Impact of Deep Learning on Tweet Ranking
As it advances deep learning models to understand user behavior, Twitter will conduct more tests to deliver the most relevant tweets to its users.
On tweet ranking in timelines, Twitter says deep learning is paying off with substantial gains in accuracy tests. Notably, its online experiments have shown increases in time spent on the platform and tweet engagement.
Before ranking, Twitter conceived of a unified, fluid, and fast model that blends easily with available techniques and modules for deep learning.
Koumchatzky says this flexibility allowed rapid testing on tweet ranking in timelines, leading to more superior models.
Due to their success, he adds that more teams at Twitter are integrating deep learning into their models.
Future of Algorithms in Social Platforms
Before tweet ranking, Twitter faced stagnant growth and waning engagement.
In the end, Twitter’s revelation proves what has been proven–algorithms work. And Twitter now has a better understanding of each tweet and the relevance of real-time interactions.
As a user, you will see more relevant tweets at the top of your timeline. As a marketer, you will have more opportunities to tap into your tweets and interactions.
Image credit: justgrimes(Flickr CC by SA 2.0)
