Elon Musk kept his promise and made available a part of the Twitter code that contains the "algorithm" for recommending content that goes into the "For You" section. Obviously, for security reasons, the parts that manage the exclusions of tweets for spam and other violations are missing. Six years after my previous article on the topic , let's see what has changed thanks to the notes of Aakash Gupta , Steven Tey and Twitter itself, who analyzed the code in detail. The process of choosing tweets As with every social media, Twitter also tries to answer the question "what content might interest each individual user the most?" The technological system that tries to answer this question consists of three phases: The collection of a series of tweets (around 1500) candidates for inclusion in the "For You" feed ( candidate sourcing ); The sorting of each individual tweet according to a score The application of filters to exclude tweets that do not comply with the policies or that the user has already seen/blocked.


Tweets outside our network (made up of people we follow) are India Mobile Number Data also excluded unless someone we follow has interacted with the tweet or is a follower of the author of the tweet. The candidate tweets are chosen 50% from those made by people who follow each other and 50% from people who do not follow each other. So Twitter groups us into clusters of similar accounts for the topics covered. The consequence is that if we make a tweet that is not related to the topics covered by the cluster to which we belong, we risk being penalised. But what affects the scoring? The reputation of the account and the weights attributed to each tweet. The largest clusters of Twitter users Tweepcred: the algorithm that calculates reputation on Twitter Twitter's algorithm called Tweepcred is derived from Google's PageRank and works like this: A score is assigned to each user based on the quantity and quality of interactions they generate during their activities on the platform A reputation score is calculated based on factors such as account longevity, number of followers and usage The score is adjusted based on the follower/following ratio The final score, on a scale from 0 to 100, represents the Tweepcred score which indicates the overall reputation on Twitter.



It is used to determine which users to recommend to follow or which users to reward with greater content visibility. The weights of the content visibility algorithm on Twitter The score to be attributed to each candidate tweet is established by respecting weights (which could change in the future): Users who follow many people, but who are not reciprocated, are penalized. When the following/follower ratio is unbalanced, the chances of being selected by the algorithm are very low; The positive factors that influence the selection are the number of likes, retweets and replies obtained, but also the time spent viewing the tweet. In particular, the weights are as follows: like: 0.5 points retweet: 1 point when someone clicks on the tweet and replies/likes the tweet or stays there for over two minutes: 11 points if someone visits your profile and leaves a like or replies to your tweet: 12 points if someone replies to your tweet: 27 points if someone replies to your tweet and you reply: 75 points The negative and therefore penalizing factors are: when someone clicks on “show less” or blocks or mutes you: -74 points when someone reports your tweet: -369 points Tweets with images and videos score twice as high as those without Recent tweets are preferred by the algorithm. A tweet's relevance score decreases by 50% every 6 hours.