关于「两种来源:推荐脉冲、搜索地板与偏短的早期半衰期」,《长视频半衰期与深度信任》给出的定量结论是什么?

结论:第二章拆开推荐脉冲与搜索地板,说明只看头一个月会把三年总量估成真值的 0.512。

数据口径:示意曲线下,短视频头 48 小时拿到三年观看的 0.750,带搜索地板的长视频只拿到 0.059,第 1,095 天仍有每天 5.00 次观看;半衰期翻十倍时,短内容峰值要高出 9.96 倍才打平现值。 书中参数为示意,全部计算由书中程序生成。

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: 0.512

A. monthviewssearch sharelocal half-life (days)
113,8600.08512.8
22,6540.42127.7
31,2400.850106.4
68861.000362.9
126261.000365.0
243131.000365.0
351661.000365.0
B. one exponential through days 0 and 30: half-life 11.2 d
predicted 36-month total 16,824true 32,857ratio 0.512
查看完整验算程序
# Chapter 2. p21: one video, two traffic sources, a half-life that grows.
# Curve Q of p11: a push burst 1000 * 2^(-t/10) (home feed, subscribers)
# plus a search floor 40 * 2^(-t/365). For each month we print the views,
# the search share, and the "local half-life" an analyst would read off
# two neighbouring months: h = 30.4375 ln 2 / ln(V_m / V_{m+1}) days.
# Part B shows the error of fitting one exponential to the first 30 days.
# All parameters are illustrative.
import math

DPM = 30.4375
PUSH, SEARCH = (1000.0, 10.0), (40.0, 365.0)


def cum1(c, t0, t1):
    a, h = c
    k = math.log(2.0) / h
    return a / k * (math.exp(-k * t0) - math.exp(-k * t1))


def month(c, m):
    return cum1(c, (m - 1) * DPM, m * DPM)


if __name__ == "__main__":
    print("A. month   views   search share   local half-life (days)")
    for m in (1, 2, 3, 6, 12, 24, 35):
        v = month(PUSH, m) + month(SEARCH, m)
        v2 = month(PUSH, m + 1) + month(SEARCH, m + 1)
        share = month(SEARCH, m) / v
        hl = DPM * math.log(2.0) / math.log(v / v2)
        print("    {:>3}   {:>6,.0f}     {:.3f}         {:>7.1f}".format(
            m, v, share, hl))
    tot = math.fsum([cum1(PUSH, 0.0, 1095.0), cum1(SEARCH, 0.0, 1095.0)])
    # a single exponential fitted to days 0 and 30 of the total curve
    r0 = PUSH[0] + SEARCH[0]
    r30 = math.fsum([PUSH[0] * 2.0 ** (-30.0 / PUSH[1]),
                     SEARCH[0] * 2.0 ** (-30.0 / SEARCH[1])])
    h_fit = 30.0 * math.log(2.0) / math.log(r0 / r30)
    naive = r0 * h_fit / math.log(2.0) * (1 - 2.0 ** (-1095.0 / h_fit))
    print("B. one exponential through days 0 and 30: half-life {:.1f} d"
          .format(h_fit))
    print("   predicted 36-month total {:,.0f}  true {:,.0f}  ratio {:.3f}"
          .format(naive, tot, naive / tot))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《长视频搜索半衰期红利、深度信任预售与多年期被动获客资产母典》第 2 章「两种来源:推荐脉冲、搜索地板与偏短的早期半衰期」(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-370 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026370,
  author    = {Fan, Yafu},
  title     = {Long-Form Video Search Half-Lives, Deep-Trust Pre-Selling and Multi-Year Passive Acquisition Assets},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论