Twitter is one of the most used platform in the social media world as its algorithm plays a huge role in what the users see on their timeline. What Twitter’s old algorithm used to be has seen a big shift and now things are different on the app. Previously the old-fashioned chronological timeline, Twitter’s algorithm consisted of Showed fewer external links, Elevated lots of “suggested” tweets (from non-followed accounts), which showed a greater diversity of sources, slightly shifted exposure to different topics, and had a slight partisan “echo chamber” effect while the new algorithm system has helped in the application in several more ways and has directed the attention of millions of users along with attracting new users every day.
If you are an old Twitter user, you might remember the uproar that took place regarding the old Twitter algorithm with people using hashtags like “RIPTwitter” and “Algorithm ruins everything” in order to show their disdain to the working of the old algorithm and now when the new shift in algorithm has taken place there is only a little the company has enlightened the people with.
Twitter has disclosed that they use a deep learning system, internally called “DeepBird,” to predict which tweets users will find interesting and engaging and since this is the limited information the platform decided to provide and deep learning systems are notorious for being “black boxes” Jack Brandy and his advisor Dr. Nicholas Diakopoulos worked together to learn more about the secrets that the Twitter’s new algorithm holds and tested Twitter’s timeline algorithm using a group of automated “puppet accounts,” comparing the puppets’ chronological timelines (“latest tweets”) to their algorithmic timelines (“top tweets”) and safe to say their findings were rather interesting.
One main aspect of the study that was observed was fewer external links. Twitter in its curation algorithm reduced the exposure to the external links. It was noticed that in the chronological timelines 51 percent of tweets contained external links while in algorithmic timelines only 18 percent did. At the same time, the exposure rate of the internal Twitter links had increased from 12 to 13 percent and the exposure rate to internal pictures increased from 19% to 30%.
It was also noticed that the timeline consisted a lot of suggested Tweets, which means that tweets which come from non-followed accounts. Overall 55 percent of the algorithmic timeline consisted of non-followed account tweets and that means the tweets from followed accounts only appeared to be less than half.
Increased source diversity was also observed as the number of unique accounts in the algorithm almost doubled from 663 in the average chronological timeline to 1,169 in the average algorithmic timeline.
Accounts that were top in tweeting also reined the algorithm system with the ten most-tweeting accounts made up 52% of tweets in the chronological timeline, but just 24% of tweets in the algorithmic timeline. According to the researchers this evidence denies the filter bubble concept people hold from social media companies and from their data they have concluded that instead of trapping users in a bubble of sources, our evidence suggests that Twitter’s algorithm diversifies the timeline with accounts that would not appear in a chronological timeline, and also reins in accounts that would dominate chronological timelines.
Next up, in order to check how the timeline algorithm in Twitter may shift the topical makeup of user timelines four tweets were observed that were political information that was about the president’s response to the pandemic, health information that included risk factors, a cluster containing economic information which was about GDP and Job loss, a cluster about fatalities e.g. death toll reports that occurred due to the pandemic.
In this survey Twitter reduced exposure to all the topics except the political cluster one and they concluded that social media algorithm sometimes reduces exposure to certain topics while elevate other topics at the same time and this factor can be called the “echo chamber” where some topics became louder while others were drowned out
The next in testing was to see that whether or not Twitter’s algorithm changed exposure to accounts with different political leanings and this was not to test that if twitter was politically biased or not but to see how the algorithm affected exposure to partisan accounts compared to users’ chronological timelines. In this testing a slight echo chamber was observed as the algorithmic timeline decreased exposure to accounts that were classified as bipartisan. In this 43% of the chronological timelines came from bipartisan accounts for left leaning puppets which decreased to 22% in the algorithmic timelines.
While Twitter has never given a through report on how its algorithm system works this is the first ever report that provides an empirical, data-backed characterization of how Twitter’s timeline algorithm changes what we see on the platform.
Read next: Twitter Survey Reveals User Optimism About Post Pandemic World
If you are an old Twitter user, you might remember the uproar that took place regarding the old Twitter algorithm with people using hashtags like “RIPTwitter” and “Algorithm ruins everything” in order to show their disdain to the working of the old algorithm and now when the new shift in algorithm has taken place there is only a little the company has enlightened the people with.
Twitter has disclosed that they use a deep learning system, internally called “DeepBird,” to predict which tweets users will find interesting and engaging and since this is the limited information the platform decided to provide and deep learning systems are notorious for being “black boxes” Jack Brandy and his advisor Dr. Nicholas Diakopoulos worked together to learn more about the secrets that the Twitter’s new algorithm holds and tested Twitter’s timeline algorithm using a group of automated “puppet accounts,” comparing the puppets’ chronological timelines (“latest tweets”) to their algorithmic timelines (“top tweets”) and safe to say their findings were rather interesting.
One main aspect of the study that was observed was fewer external links. Twitter in its curation algorithm reduced the exposure to the external links. It was noticed that in the chronological timelines 51 percent of tweets contained external links while in algorithmic timelines only 18 percent did. At the same time, the exposure rate of the internal Twitter links had increased from 12 to 13 percent and the exposure rate to internal pictures increased from 19% to 30%.
It was also noticed that the timeline consisted a lot of suggested Tweets, which means that tweets which come from non-followed accounts. Overall 55 percent of the algorithmic timeline consisted of non-followed account tweets and that means the tweets from followed accounts only appeared to be less than half.
Increased source diversity was also observed as the number of unique accounts in the algorithm almost doubled from 663 in the average chronological timeline to 1,169 in the average algorithmic timeline.
Accounts that were top in tweeting also reined the algorithm system with the ten most-tweeting accounts made up 52% of tweets in the chronological timeline, but just 24% of tweets in the algorithmic timeline. According to the researchers this evidence denies the filter bubble concept people hold from social media companies and from their data they have concluded that instead of trapping users in a bubble of sources, our evidence suggests that Twitter’s algorithm diversifies the timeline with accounts that would not appear in a chronological timeline, and also reins in accounts that would dominate chronological timelines.
Next up, in order to check how the timeline algorithm in Twitter may shift the topical makeup of user timelines four tweets were observed that were political information that was about the president’s response to the pandemic, health information that included risk factors, a cluster containing economic information which was about GDP and Job loss, a cluster about fatalities e.g. death toll reports that occurred due to the pandemic.
In this survey Twitter reduced exposure to all the topics except the political cluster one and they concluded that social media algorithm sometimes reduces exposure to certain topics while elevate other topics at the same time and this factor can be called the “echo chamber” where some topics became louder while others were drowned out
The next in testing was to see that whether or not Twitter’s algorithm changed exposure to accounts with different political leanings and this was not to test that if twitter was politically biased or not but to see how the algorithm affected exposure to partisan accounts compared to users’ chronological timelines. In this testing a slight echo chamber was observed as the algorithmic timeline decreased exposure to accounts that were classified as bipartisan. In this 43% of the chronological timelines came from bipartisan accounts for left leaning puppets which decreased to 22% in the algorithmic timelines.
While Twitter has never given a through report on how its algorithm system works this is the first ever report that provides an empirical, data-backed characterization of how Twitter’s timeline algorithm changes what we see on the platform.
Read next: Twitter Survey Reveals User Optimism About Post Pandemic World
