关于「Threads 与 X:回复链、回复权重、早期速度与时间衰减」,《五大平台算法与一鱼五吃》给出的定量结论是什么?

结论:第三章说明负面反馈概率超过 0.0020 时挑衅帖子在公开权重下变成负分,最好与最差发帖时刻相差 20.0 倍。

数据口径:书中把每条平台说法标为三类:带日期的官方披露、同行评审研究、本书的示意模型;书中的权重、概率与漏斗参数均为示意,不是任何平台的真实设定,全部计算由书中程序生成。 本书是一般性经营参考,不构成法律、税务或投资建议;不涉及假账号、机器人、互赞群、买互动、抓取个人资料或群发私讯;合作内容一律标明;达人酬劳只有单层直付,不涉及任何多层分润。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 0.0020

agreeable post score0.0408
question + replies score0.1277
rage post score-0.2978
rage post turns negative once P(negative feedback) > 0.0020
sigma 0.5, share of draws ranked first: agreeable post 0.034, question + replies 0.962, rage post 0.005
sigma 1.0, share of draws ranked first: agreeable post 0.214, question + replies 0.718, rage post 0.069
查看完整验算程序
# Chapter 3. p31: three post styles under the weights X published on 5 April 2023,
# the negative-feedback level that turns a rage post negative, and robustness when
# every weight is perturbed. Weights: official 2023 disclosure (the-algorithm-ml,
# projects/home/recap/README.md). All per-viewer probabilities are illustrative.
# Perturbation: each weight times exp(N(0, sigma)), sigma 0.5 and 1.0, 2,000 seeded
# draws each; signs are kept.
import math
import random

RNG = random.Random(359)
W = {"fav": 0.5, "retweet": 1.0, "reply": 13.5, "areply": 75.0, "neg": -74.0, "report": -369.0}
STYLE = {
    "agreeable post": {"fav": 0.060, "retweet": 0.012, "reply": 0.0010, "areply": 0.0001,
                       "neg": 0.0002, "report": 0.00002},
    "question + replies": {"fav": 0.020, "retweet": 0.003, "reply": 0.0040, "areply": 0.0015,
                           "neg": 0.0006, "report": 0.00002},
    "rage post": {"fav": 0.025, "retweet": 0.008, "reply": 0.0120, "areply": 0.0005,
                  "neg": 0.0060, "report": 0.00020},
}


def score(w, p):
    return math.fsum(w[k] * p[k] for k in w)


for name, p in STYLE.items():
    print("%-19s score %8.4f" % (name, score(W, p)))
rage = STYLE["rage post"]
pos = math.fsum(W[k] * rage[k] for k in ("fav", "retweet", "reply", "areply"))
rep = W["report"] * rage["report"]
print("rage post turns negative once P(negative feedback) > %.4f" % ((pos + rep) / -W["neg"]))
for sigma in (0.5, 1.0):
    first = {n: 0 for n in STYLE}
    for _ in range(2000):
        w = {k: v * math.exp(RNG.gauss(0.0, sigma)) for k, v in W.items()}
        first[max(STYLE, key=lambda n: score(w, STYLE[n]))] += 1
    print("sigma %.1f, share of draws ranked first: " % sigma + ", ".join(
        "%s %.3f" % (n, c / 2000) for n, c in first.items()))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《五大社交图谱算法异质性、跨平台同源异构分发与达人聚拢母典》第 3 章「Threads 与 X:回复链、回复权重、早期速度与时间衰减」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-359 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026359,
  author    = {Fan, Yafu},
  title     = {Algorithmic Heterogeneity across RedNote, Threads, X, Instagram and Facebook: Cross-Platform Adaptation and Creator Aggregation},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论