关于「三秒钩子,生存分析」,《暗投授权与钩子归因》给出的定量结论是什么?

结论:第三章用卡普兰与迈耶曲线、威布尔拟合分析三秒钩子,形状参数 0.582,留住订单在钩子强度 0.26 时最高。

数据口径:书中参数均为示意,全部计算由书中程序生成;本书是非个人化经营分析,不构成经营或法律建议。 第六章用二叉树为跳单期权定价,合理预估 20.82 已拦住示意模型中全部跳单。

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: 0.582

all views: shape k0.582, scale = 13.47 s
S(3) fitted 0.6586
hazard at0.5 s: 0.1713 per second
hazard at1.0 s: 0.1282 per second
hazard at3.0 s: 0.0809 per second
hazard at 10.0 s0.0489 per second
hazard at 20.0 s0.0366 per second
hazard ratio 0.5 s vs 10 s3.50
after 3 s: shape k1.009, scale = 25.22 s, n = 1,192
查看完整验算程序
# Chapter 3. p32: Weibull fit with censoring; shape below 1 = early spike.
# Same synthetic views as p31 (same generator and seed). Maximum
# likelihood for Weibull(scale lam, shape k) with right censoring:
# the profile score in k,
#   1/k + mean(ln t over events) - sum(t^k ln t) / sum(t^k) = 0,
# is solved by bisection; then lam^k = sum(t^k) / events.
# Hazard h(t) = (k / lam) (t / lam)^(k - 1). A second fit uses only
# views that passed 3 s, on the time left after 3 s.
import math
import random

N, LENGTH = 2000, 30.0


def views():
    rng = random.Random(3330)
    out = []
    for _ in range(N):
        if rng.random() < 0.42:
            t = rng.weibullvariate(1.1, 0.9)
        else:
            t = rng.weibullvariate(28.0, 1.3)
        event = True
        if rng.random() < 0.05:
            cut = rng.uniform(0.0, LENGTH)
            if cut < t:
                t, event = cut, False
        if t >= LENGTH:
            t, event = LENGTH, False
        t = max(0.1, math.ceil(t * 10.0) / 10.0)
        out.append((t, event))
    return out


def fit(data):
    ev = [math.log(t) for t, e in data if e]
    d, mlog = len(ev), math.fsum(ev) / len(ev)

    def score(k):
        tk = [t ** k for t, e in data]
        tkl = [t ** k * math.log(t) for t, e in data]
        return 1.0 / k + mlog - math.fsum(tkl) / math.fsum(tk)

    lo, hi = 0.05, 5.0
    for _ in range(80):
        mid = 0.5 * (lo + hi)
        if score(mid) > 0:
            lo = mid
        else:
            hi = mid
    k = 0.5 * (lo + hi)
    lam = (math.fsum(t ** k for t, e in data) / d) ** (1.0 / k)
    return k, lam


def haz(t, k, lam):
    return k / lam * (t / lam) ** (k - 1.0)


if __name__ == "__main__":
    data = views()
    k, lam = fit(data)
    print("all views:   shape k = {:.3f}, scale = {:.2f} s".format(k, lam))
    print("  S(3) fitted {:.4f}".format(math.exp(-(3.0 / lam) ** k)))
    for t in (0.5, 1.0, 3.0, 10.0, 20.0):
        print("  hazard at {:>4.1f} s: {:.4f} per second".format(
            t, haz(t, k, lam)))
    print("  hazard ratio 0.5 s vs 10 s: {:.2f}".format(
        haz(0.5, k, lam) / haz(10.0, k, lam)))
    late = [(round(t - 3.0, 1), e) for t, e in data if t > 3.0]
    k2, lam2 = fit(late)
    print("after 3 s:   shape k = {:.3f}, scale = {:.2f} s, n = {:,}"
          .format(k2, lam2, len(late)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《暗投白名单授权经济学、三秒钩子生存分析与单层星探归因博弈母典》第 3 章「三秒钩子,生存分析」(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-333 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026333,
  author    = {Fan, Yafu},
  title     = {Dark-Ads Whitelist Licensing Economics, Three-Second Hook Survival Analysis and Single-Tier Scout Attribution Games},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论