关于「单层星探,沙普利分成」,《暗投授权与钩子归因》给出的定量结论是什么?
结论:第五章以沙普利值做单层星探分成,星探份额 0.1458,末次触点下广告主净得 0.2000,沙普利值下升到 0.3561,零贡献触点分成为零。
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。
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。 与结论吻合的打印值: 0.2000, 0.3561
| rule | x1 | x2 | P | lift | adv. net | welfare |
| last touch | 0.000 | 1.000 | 0.4000 | 0.1000 | 0.2000 | 0.3500 |
| shapley | 0.377 | 0.453 | 0.4122 | 0.1122 | 0.3561 | 0.3841 |
| efficient | 0.670 | 0.670 | 0.4786 | 0.1786 | 0.4786 | 0.4000 |
查看完整验算程序
# Chapter 5. p52: last touch rewards exposure, Shapley rewards lift.
# Two touch points on one target cohort: x1 = reach of the licensed
# creator ad (upper funnel), x2 = intensity of retargeting (shown
# last). Purchase probability P = p0 + a1 x1 + a2 x2 - a12 x1 x2 with
# p0 = 0.30 (would buy anyway), a1 = 0.20, a2 = 0.10, a12 = 0.05.
# Party i pays k_i x_i^2 / 2 (k1 = 0.25, k2 = 0.10) and earns r = 0.5
# per credited order. Credit rules:
# last touch: party 2 gets x2 P, party 1 gets x1 (1 - x2) P
# Shapley on lift: party 1 gets a1 x1 - a12 x1 x2 / 2, party 2
# gets a2 x2 - a12 x1 x2 / 2; baseline orders go to nobody.
# Nash equilibrium by best-response iteration on a 0.001 grid; the
# advertiser earns margin 1 per order minus credits paid. A stylised
# reading of Berman (2018); parameters illustrative.
P0, A1, A2, A12, K1, K2, R = 0.30, 0.20, 0.10, 0.05, 0.25, 0.10, 0.5
GRID = [j / 1000.0 for j in range(1001)]
def prob(x1, x2):
return P0 + A1 * x1 + A2 * x2 - A12 * x1 * x2
def credit(rule, x1, x2):
if rule == "last touch":
p = prob(x1, x2)
return x1 * (1.0 - x2) * p, x2 * p
return (A1 * x1 - A12 * x1 * x2 / 2.0,
A2 * x2 - A12 * x1 * x2 / 2.0)
def equilibrium(rule):
x1, x2 = 0.5, 0.5
for _ in range(200):
n1 = max(GRID, key=lambda y: R * credit(rule, y, x2)[0]
- K1 * y * y / 2.0)
n2 = max(GRID, key=lambda y: R * credit(rule, n1, y)[1]
- K2 * y * y / 2.0)
if (n1, n2) == (x1, x2):
break
x1, x2 = n1, n2
return x1, x2
def report(name, x1, x2, paid):
p = prob(x1, x2)
cost = K1 * x1 * x1 / 2.0 + K2 * x2 * x2 / 2.0
print(" {:<10} {:.3f} {:.3f} {:.4f} {:.4f} {:>9.4f} {:.4f}"
.format(name, x1, x2, p, p - P0, p - paid, p - cost))
if __name__ == "__main__":
print(" rule x1 x2 P lift adv. net"
" welfare")
for rule in ("last touch", "shapley"):
x1, x2 = equilibrium(rule)
c1, c2 = credit(rule, x1, x2)
report(rule, x1, x2, R * (c1 + c2))
best = max(((a, b) for a in GRID[::10] for b in GRID[::10]),
key=lambda z: prob(*z) - K1 * z[0] ** 2 / 2
- K2 * z[1] ** 2 / 2)
report("efficient", best[0], best[1], 0.0)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 5 章「单层星探,沙普利分成」(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}
}