Musk is such a man. When he said he would open source X/Twitter's algorithm, he really opened it up. In terms of military strategy, knowing oneself and one's enemy ensures victory in a hundred battles. Whether you are a reader, author, or both, you should read the interpretation and analysis of Twitter's algorithm, which determines (1) what you can brush up on? (2) How can you express your views so that more people can see them? 1、 Some common sense 1.1 Concept of Internal and External Networks Twitter has introduced a fundamental concept: Thunder represents the account you are currently following on the intranet; Phoenix Retrieval represents the entire Twitter content, including accounts you haven't followed. 1.2 Introduction of Gork analysis Everyone has a hidden Grok assistant. As an author, Grok mini will first conduct an analysis and predict user engagement with the content you write. As a reader, Grok is predicting what posts you want to see and then determining whether to search for posts on the Thunderbolt intranet or the Phoenix search intranet. An example is that if you like and click on an article, it will push various articles to you. If you don't like to read the clue posts, stop displaying them. If you love videos, I'll recommend them to you. For comparison, Twitter in 2023 has a hard coded rule, where if a post has a video, the score is multiplied by 2. If there is a link, divide the score by 2. This is no longer a mechanical scoring system. 2、 Some rules that people may overlook 2.1 Invisible Credit Score A legacy algorithm (still running in the filter) assigns a reputation score to each user, ranging from -128 to+100. The initial credit of the new account is -128, in other words, the default newcomers are all bad guys (want to roast about the same as HSBC...). Before you gain positive reputation through continuous interaction with high reputation users, posts are usually "soft banned" or restricted from spreading. Additionally, if you frequently interact with low-quality/junk accounts (such as AI accounts or Huangtu Ge), your reputation score will not be high. 2.2 Topic fatigue The system will actively reduce the ranking of multiple tweets from the same author in a short period of time. If you are the author, don't post too many articles in an hour, just 5 or 10. The system also knows that one person cannot have so much practical knowledge in a short period of time. Content redundancy: Repetitive and low input text on popular topics will be penalized. The system will detect 'topic fatigue' - if a user has seen 10 posts about Vibe Coding today, even if the quality is good, the 11th post will be significantly downgraded. So, when it comes to hot topics, they should be posted early; if they are too late, they will be useless. 2.3 High weight of negative feedback The algorithm has a much higher weight for "displaying less similar content", muting, and blacking out than for positive likes, with a significant difference between the two. So, if you always play extreme games, such as indiscriminate attacks and mocking a certain group, you will eventually cool down. One blacklist equals a bunch of die hard fans. The neighboring fan circle has long said that one black equals one hundred fans, which is probably what it means. In addition, deleting/spamming/violent/bloody/etc. are all low scores, as we all know. PS: But strangely, it doesn't seem to mention pornographic content. Does Musk still encourage everyone to watch H?? 2.4 Blue V is very important The input of the model is largely biased towards users with verified accounts (Premium or Blue V), which are easier to be retrieved by Phoenix Search's external network. Retrieval has strict thresholds, and unverified accounts typically cannot enter the initial candidate pool unless their participation grows rapidly. That is to say, if you are not a blue V, it will mainly be distributed to your existing fans; You are a blue V, so you can regularly distribute to all Twitter fans. 3、 Other small points -Even if users don't like or comment, and just stare at it normally (dwell), they will still receive extra points; -Candidate quarantine. The algorithm will analyze tweets in isolation, without affecting the score of a single post due to a batch of posts. Let's talk about candidate isolation here. Imagine you are a book buyer. When the bookstore owner recommends books to you, if you simply score them, they may recommend all best-selling books to you. This is obviously not right, because you have your own preferences. Under the candidate isolation rule, the algorithm will guess what you like to watch for you, rather than simply pushing you based on your score. -Algorithms do not like duplicate content or old posts -The importance of completion rate has increased when posting videos. That's all for now, welcome to discuss in the comments section what I haven't mentioned.