Twitter Ranking Factors & Process 2023

Since Twitter open-sourced its ranking algorithm on Mar 31 , 2023, Here are its ranking factors and their ranking process in detail

Twitter Ranking Process

1. From 200 Million tweets every day 1500 tweets are identified for the for you feed for a user
2. 50% of tweets come from users which are being followed and 50% of tweets come from recommendation
3. Twitter uses a Heavy ranker which uses regression Machine Learning to suggest tweets that are relevant to a user and have a higher potential for engagement
4. The Machine learning suggested feed is modified manually using filters and presented to the user

How Twitter Identifies 1500 tweets

For the first 1500 tweets, Twitter sources 50% for users whom you are following and tries to identify recent and relevant tweets this is how it does so

In Network Source

From the users whom you follow Twitter identifies relevant tweets based on the Real graph model. This is a logistic regression model which tries to predict the likelihood of engagement between you and the user whom you are following and the higher the chance of engagement more the score more such tweets will be included

Twitter also sources 50% of tweets from users whom you don't follow and this is how it does so

Out of Network Sources

Twitter tries to calculate the relevance of tweets by matching the engagements of people whom you follow and those with similar interests. Twitter also calculates the similarity between users, tweets, and user tweets pair using a vector approach called Sim Clusters

Twitter Also uses PageRank to calculate a user's reputation 

How Twitter Heavy Ranker  Ranks Tweets

Twitter Heavy Ranker ranks tweets by generating a probability and a weight for each engagement

This is how the final score is calculated.

score = sum_i { (weight of engagement i) * (probability of engagement i) }

a) This is how probability is generated

scored_tweets_model_weight_fav: The probability the user will favorite the Tweet. scored_tweets_model_weight_retweet: The probability the user will Retweet the Tweet. scored_tweets_model_weight_reply: The probability the user replies to the Tweet. scored_tweets_model_weight_good_profile_click: The probability the user opens the Tweet author profile and Likes or replies to a Tweet. 

scored_tweets_model_weight_video_playback50: The probability (for a video Tweet) that the user will watch at least half of the video. 

scored_tweets_model_weight_reply_engaged_by_author: The probability the user replies to the Tweet and this reply is engaged by the Tweet author.

 scored_tweets_model_weight_good_click: The probability the user will click into the conversation of this Tweet and reply or Like a Tweet. 

scored_tweets_model_weight_good_click_v2: The probability the user will click into the conversation of this Tweet and stay there for at least 2 minutes. scored_tweets_model_weight_negative_feedback_v2: The probability the user will react negatively (requesting "show less often" on the Tweet or author, block or mute the Tweet author). scored_tweets_model_weight_report: The probability the user will click Report Tweet.

b) And these are the weights it uses as of April 5, 2023

scored_tweets_model_weight_fav: 0.5 

scored_tweets_model_weight_retweet: 1.0 

scored_tweets_model_weight_reply: 13.5 

scored_tweets_model_weight_good_profile_click: 12.0 scored_tweets_model_weight_video_playback50: 0.005 scored_tweets_model_weight_reply_engaged_by_author: 75.0 scored_tweets_model_weight_good_click: 11.0 

scored_tweets_model_weight_good_click_v2: 10.0 scored_tweets_model_weight_negative_feedback_v2: -74.0

 scored_tweets_model_weight_report: -369.0

Looks like it weighs replies most followed by profile click, and the fact that the reply is engaged by the tweet author and then good click i.e stay for 2 minutes or more

Reference: https://github.com/twitter/the-algorithm-ml/tree/main/projects/home/recap

Earlier Analysis (How Twitter Ranks tweets)

Twitter uses a Machine learning algorithm that optimizes for engagement and it focuses on likes first then retweets then english language tweets and then boosts images and videos and deprioritizes links ( This is based on light ranker but it seems twitter uses heavy ranker)

These are the ranking factors

1.  Number of Likes  weight 30 the highest
2. Number of Retweets weight 20
3. Indirect follows weight 4  ( You follow someone who follows the author of the tweet )
4. In trusted Circle weight 3 ( based on the similarity scores you are in their trusted circle)
5. Link weight 2 ( Negative ranking factor unless you are a verified user or you get enough engagement to overcome it you will get marked as spam)
6. Self tweets weight 2 ( Boosts tweets in which you reply to your own tweets i.e threads)
7. Tweets has image weight 2 boosts tweets having images
8. Tweets has video weight 2 boosts tweets having videos
9. Tweets has trend weight 1.1 boosts tweets talking about a trending topic
10.  Tweet has a reply has a weight of 1
11. tweet has a reply count has a weight of 1 ( boosting tweets having a lot of reply)
12. Multiple hashtag weight 0.6 ( negative boost for a tweet having multiple hashtags)
13. reputation boost weight 0.2 ( measures reputation of user using pagerank and uses that for positive or negative boosts small impact)
14. English language boost weight 0.5
15. offensive negative weight 0.1 for tweets that are marked offensive
16. boost for twitter blue and verified accounts

How Twitter Filters tweets

Twitter filters tweets  based on the following factors

1. Blocked and muted accounts are removed
2. Author diversity is improved so that you don't see all tweets from the same account
3. Content Balance filter to maintain a balance of in-network (people you follow) and out-of-network (people you don't follow )
4. Negative Feedback tweets are lowered
5. Social proof  ( A filter to ensure that you only see tweets from people with whom you have a connection)
6. Conversation Filter ( replies to a thread are grouped together

Other filters like NSFW, DMCA , Toxicity, Ukraine, Misinformation, legal, Coordinated activity,Hate, violence

Twitter Takeaway for Creators/ Publishers

Primarily Twitter ranking is optimizing for engagement on the platform and has a negative weight for links which means that you cannot get traffic from Twitter as a publisher.

As a creator, you should optimize your tweets for likes and retweets more than replies. The use of threads will provide a boost as well as using images and videos and English. Live and die on the platform Here is the algorithm in the pic provided by twitter