Twitter Updates Recommendation Algorithm Weights Under Elon Musk’s Leadership: An Analysis
Earlier, when Twitter released the source code of its recommendation algorithm, users analyzed the code to determine the weights assigned to tweets being recommended. For instance, a tweet’s “like” weight was found to be higher than its retweet weight.
However, Twitter’s engineering team has announced that the previously disclosed algorithm weights were a legacy from the Jack Dorsey era. The recommendation algorithm has since been updated under Elon Musk’s leadership, and the associated weights have also been adjusted. To avoid confusion, Twitter has replaced the original weight portion on GitHub with the new weight model.
These changes are still entirely public, allowing anyone to examine and analyze the source code of Twitter’s algorithm. The weight portion is described separately, making it easy to analyze.
Here are the new weights for Twitter’s recommendation algorithm:
- scored_tweets_model_weight_fav: 0.5 — The model score weight when a tweet is marked as “liked.”
- scored_tweets_model_weight_retweet: 1.0 — The model score weight when a tweet is retweeted.
- scored_tweets_model_weight_reply: 13.5 — The model score weight when a tweet receives a reply.
- scored_tweets_model_weight_good_profile_click: 12.0 — The model score weight when a tweet leads to a user clicking on another profile, considered a positive action.
- scored_tweets_model_weight_video_playback50: 0.005 — The model score weight when a video in a tweet is played for at least 50%.
- scored_tweets_model_weight_reply_engaged_by_author: 75.0 — The model score weight when the tweet author engages in the reply discussion.
- scored_tweets_model_weight_good_click: 11.0 — The model score weight when a link in a tweet is considered a positive click.
- scored_tweets_model_weight_good_click_v2: 10.0 — Similar to the previous weight, but this is the second version, with a weight of 10.0.
- scored_tweets_model_weight_negative_feedback_v2: -74.0 — The model score weight when a tweet receives negative feedback (e.g., dislikes or dissatisfaction) is -74.0.
- scored_tweets_model_weight_report: -369.0 — The model score weight when a tweet is reported is -369.0.
It’s worth noting that Elon Musk has previously hinted that the “For You” section will only recommend tweets from verified users, meaning those who have subscribed to Twitter Blue. Unverified users’ tweets will not appear in the recommendations.
The new algorithm doesn’t mention any weight for verified users. Considering Musk’s statement, it seems he wasn’t joking, as there would be no need to assign separate weights for verified users if only their tweets were being recommended in the first place.