刚刚！马斯克把 X 推荐算法底裤给开源了

Core Viewpoint - The company has announced the open-sourcing of its recommendation algorithm for X, aiming to enhance transparency and invite community collaboration in its optimization efforts [1]. Group 1: Algorithm Overview - The recommendation algorithm consists of components such as Home Mixer, Thunder, and Phoenix, focusing on sorting information from "followed accounts" and "content mined from the web" using a Grok-based Transformer model [3]. - A notable feature of the algorithm is the elimination of all manually designed features and heuristic rules, relying solely on the model to learn correlations from user interaction history [5]. Group 2: Algorithm Functionality - The algorithm's workflow involves capturing user behavior data, sourcing candidate content from two channels, and applying data augmentation and multi-round filtering to predict user engagement probabilities (likes, shares, etc.), ultimately calculating a final score for content presentation [5]. - The implementation of X's recommendation algorithm is done using Rust and Python, and it is licensed under the Apache-2.0 license. As of now, it has received 1.1k stars and 195 forks on GitHub [5].