purplETHought #1: Twitter Algorithm

The implications of taking the Twitter algorithm open-source.

On March 31st, Twitter released the code showing how the algorithm works and ranks the over 150 billion tweets sent to people’s devices daily. In addition, the company shared both the GitHub and a blog post detailing their decision.

Twitter stated that “the goal of our open source endeavor is to provide full transparency to you, our users, about how our systems work.”

A self-fulfilling prophecy

After the release, a race started on Twitter with people looking into the algorithm and trying to understand the code and its inner workings. I used ChatGPT to analyze how tweets are ranked but also predicted that it is likely to see the algorithm change. If everyone knows how it works, everyone will optimize for the algorithm, making it redundant.

From static values to machine learning

Yesterday evening, we saw that prediction coming to life when the Twitter algorithm received a massive overhaul. Do you remember the threads about weighting and how likes are more critical than retweets? They are all outdated.

Instead of using fixed weights that specify the value of a like, comment, or retweet, the algorithm is now based on TensorFlow. Simply put, it is a machine-learning tool that constantly changes and adapts.

What makes a good tweet?

So what variables can we look at? The “TweetTextScorer” is responsible for evaluating the quality of a tweet. I talked about it in a thread here (unfortunately, Twitter embeds don’t work right now): https://twitter.com/Legendary_NFT/status/1642256440483258369?s=20

The main take factors that matter when it comes to the quality of a Tweet are the following:

  • Readability: Ensure your tweet is easy to read and understand.

  • Length: Keep the tweet's length within an optimal range.

  • Shoutness: Avoid using too many capital letters or exclamation marks, as they might be interpreted as shouting.

  • Entropy: Keep the unpredictability or randomness of the text within an acceptable range.

  • Links: Minimize the number of links in the tweet, if possible.

  • Offensiveness: Ensure that your tweet doesn't contain offensive words.

Tip: ChatGPT can help you with these six factors. Just paste the 6 points above and your tweet and tell GPT to rate your tweet based on those criteria.

WTF is “Tweepcred”?

Another factor when it comes to ranking your tweets is “Tweepcred.” It is an analysis tool that determines your influence rank based on your interactions with other users. It uses the PageRank algorithm, just like Google search.

My main takeaway is that genuine engagement with other users is vital. If you ask a question, and people reply to your tweet, don’t ignore them but talk to them. Ultimately, social media is a two-way street and not an echo chamber. It is rude to ignore the people who take their precious time to reply to your tweets and answer your questions. It is good to see that the Twitter algorithm rewards being nice to others.