x-foryou-algorithm

Installation
SKILL.md

For You 算法推理结构 — 推导创作者行动

R — 原文 (Reading)

"We have eliminated every single hand-engineered feature and most heuristics. The Grok-based transformer does all the heavy lifting by understanding your engagement history. During transformer inference, candidates cannot attend to each other—only to the user context, making scores consistent and cacheable. Final Score = Σ (weight_i × P(action_i)). Positive actions (like, repost, share) have positive weights. Negative actions (block, mute, report) have negative weights, pushing down content the user would likely dislike."

— xAI, X推荐算法.md · Key Design Decisions / Scoring and Ranking

自译:我们已消除所有手工特征和大部分启发式规则。Grok transformer从你的互动历史中自行学习相关性。排序时候选帖互不可见,只attend用户上下文,使分数稳定可缓存。最终分=各动作概率的加权和,正面动作(点赞/转发/分享)正权重,负面动作(拉黑/静音/举报)负权重,主动压低用户不喜欢的内容。


I — 方法论骨架 (Interpretation)

X 的 For You 推荐流由一条流水线驱动:双源召回 → 水合过滤 → 多动作打分 → 多样性 → 选取。三个核心设计决策构成了理解"内容为什么被推荐"的推理骨架:

  1. 无手工特征:Grok transformer 从互动历史自行学习相关性,不靠人工定义"什么内容好"。内容是否相关由用户行为模式学习,而非预设规则。
  2. 候选隔离:打分时每条帖子独立评估,不能"看到"同 batch 其他帖子,只 attend 用户上下文。你的帖子按自身预测质量评分,不是"比同批帖子更好"——质量是唯一杠杆,非相对时机。
  3. 多动作加权:不预测单一"相关性"分数,而是预测 15+ 种动作概率(点赞/回复/转发/收藏/点击/停留/关注……),正动作正权重、负动作(拉黑/静音/举报)负权重,加权求和得最终分。
Installs
3
GitHub Stars
52
First Seen
Jul 15, 2026
x-foryou-algorithm — kangarooking/x-growth-skills