关于「重组素材有多不同:汉明距离、表现相关与共享镜头的疲劳」,《素材疲劳与正交混剪》给出的定量结论是什么?

结论:第四章区分设计距离、表现相关与共享镜头的疲劳:12 个镜头重组出的素材池,对观众只相当于两到三条全新素材,平均相对点击率 0.828。

数据口径:买量圈流行的说法是:扩量期素材衰退得快,因为观众重复曝光使点击率按可估计的半衰期衰减;把短视频拆成开头钩子、痛点、证据、行动召唤四个模块,用正交表组合,十几个镜头就能自动合成上百条低相关的测试素材,突破产能瓶颈。 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: 0.828

poolmean relative CTR over 8 exposures
1 fully distinct0.589
2 fully distinct0.761
3 fully distinct0.833
4 fully distinct0.871
6 fully distinct0.912
9 fully distinct0.940
16 fully distinct0.966
81 fully distinct0.993
L9 (9 recombined) 0.828 (between 2 and 3 distinct)
full 81 recombined 0.828 (between 2 and 3 distinct)
查看完整验算程序
# Chapter 4. p43: recombined creatives share shots, so they share fatigue
# A viewer fatigues on shots, not on file names. Exposure response =
# c0 * prod_m 0.5 ** (w_m * n_m / h), n_m = earlier views of the shot
# used in module m. Weights: hook 0.6, pain 0.2, proof 0.1, cta 0.1, so
# repeating one whole creative gives the plain half-life h. Each viewer
# sees 8 exposures drawn at random from the pool. Illustrative.
import itertools
import math
import random

rng = random.Random(40043)
W = [0.6, 0.2, 0.1, 0.1]
H = 4.0
EXPO = 8
VIEWERS = 20000


def viewer_mean(pool):
    seen = [{} for _ in range(4)]
    rs = []
    for _ in range(EXPO):
        g = pool[rng.randrange(len(pool))]
        expo = math.fsum(W[m] * seen[m].get(g[m], 0) for m in range(4))
        rs.append(0.5 ** (expo / H))
        for m in range(4):
            seen[m][g[m]] = seen[m].get(g[m], 0) + 1
    return math.fsum(rs) / EXPO


def pool_mean(pool):
    return math.fsum(viewer_mean(pool) for _ in range(VIEWERS)) / VIEWERS


def fresh(n):
    # n creatives that share no shot at all (each its own 4 shots)
    return [(k, k, k, k) for k in range(n)]


full = list(itertools.product(range(3), repeat=4))
l9 = [(i, j, (i + j) % 3, (i + 2 * j) % 3)
      for i, j in itertools.product(range(3), repeat=2)]
ref = [(n, pool_mean(fresh(n))) for n in (1, 2, 3, 4, 6, 9, 16, 81)]
print("pool                     mean relative CTR over 8 exposures")
for n, v in ref:
    print("{:>2} fully distinct        {:.3f}".format(n, v))
for name, pool in (("L9 (9 recombined)", l9), ("full 81 recombined", full)):
    v = pool_mean(pool)
    lo = max(n for n, r in ref if r <= v)
    hi = min(n for n, r in ref if r > v)
    print("{:<24} {:.3f}  (between {} and {} distinct)".format(
        name, v, lo, hi))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《投放素材频次疲劳半衰期、分镜基因正交重组与百条广告全自动混剪母典》第 4 章「重组素材有多不同:汉明距离、表现相关与共享镜头的疲劳」(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-400 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026400,
  author    = {Fan, Yafu},
  title     = {Ad-Creative Fatigue Half-Lives, Orthogonal Storyboard Recombination and Automated Permutation Rendering},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论