关于「J 形评论与一条差评的分量:评价如何影响成交,平均分为何失真」,《声誉反脆弱与受众主权》给出的定量结论是什么?
结论:第一章说明少报偏差把评价压成 J 形,页面平均分比真实体验高出约 0.410 星。
数据口径:书中参数为示意;全部计算由书中程序生成;本书是非个人化经营建议,不是法律意见。 书中以标注框区分已发表的证据、可运行的 Python 计算(22 段程序)、原典与作者的假说。
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: 0.410
| A. star true share post propensity posted share | |
| 1 | 0.018 0.30 0.054 |
| 2 | 0.118 0.08 0.095 |
| 3 | 0.325 0.04 0.132 |
| 4 | 0.356 0.07 0.252 |
| 5 | 0.184 0.25 0.466 |
| B. mean of true 3.570, mean of posted 3.981, gap +0.410 | |
| C. share of 1 or 5 stars: true 0.202, posted 0.521 | |
| D. extremity factor k scales the 1- and 5-star propensity | |
| k | 1.0: posted mean 3.981, 1-star share 0.054, 5-star share 0.466 |
| k | 1.5: posted mean 4.105, 1-star share 0.065, 5-star share 0.555 |
| k | 2.0: posted mean 4.187, 1-star share 0.071, 5-star share 0.613 |
| k | 3.0: posted mean 4.288, 1-star share 0.080, 5-star share 0.685 |
查看完整验算程序
# Chapter 1. p11: how under-reporting bends true ratings into a J shape.
# Model: buyers' experiences on a 1-5 star scale come from a latent
# normal (mean 3.6, sd 1.0, cut at 1.5 / 2.5 / 3.5 / 4.5). Each buyer
# posts with a propensity that depends on the star: extremes post more
# (the under-reporting bias of Hu, Zhang and Pavlou 2009). All numbers
# are illustrative; nothing here is measured from a real shop.
import math
from statistics import NormalDist
LATENT = NormalDist(3.6, 1.0)
CUTS = (1.5, 2.5, 3.5, 4.5)
BASE_PROP = (0.30, 0.08, 0.04, 0.07, 0.25)
def true_shares():
edges = (-math.inf,) + CUTS + (math.inf,)
out = []
for lo, hi in zip(edges[:-1], edges[1:]):
a = 0.0 if lo == -math.inf else LATENT.cdf(lo)
b = 1.0 if hi == math.inf else LATENT.cdf(hi)
out.append(b - a)
return out
def posted(shares, prop):
w = [s * p for s, p in zip(shares, prop)]
tot = math.fsum(w)
return [x / tot for x in w]
def mean(shares):
return math.fsum((i + 1) * s for i, s in enumerate(shares))
if __name__ == "__main__":
t = true_shares()
p = posted(t, BASE_PROP)
print("A. star true share post propensity posted share")
for i in range(5):
print(" {} {:.3f} {:.2f} {:.3f}".format(
i + 1, t[i], BASE_PROP[i], p[i]))
print("B. mean of true {:.3f}, mean of posted {:.3f}, gap {:+.3f}"
.format(mean(t), mean(p), mean(p) - mean(t)))
print("C. share of 1 or 5 stars: true {:.3f}, posted {:.3f}".format(
t[0] + t[4], p[0] + p[4]))
print("D. extremity factor k scales the 1- and 5-star propensity")
for k in (1.0, 1.5, 2.0, 3.0):
prop = (BASE_PROP[0] * k,) + BASE_PROP[1:4] + (BASE_PROP[4] * k,)
q = posted(t, prop)
print(" k = {:.1f}: posted mean {:.3f}, 1-star share {:.3f}, "
"5-star share {:.3f}".format(k, mean(q), q[0], q[4]))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《商业声誉反脆弱柔道、恶意差评证据链反制与去平台依赖私域主权母典》第 1 章「J 形评论与一条差评的分量:评价如何影响成交,平均分为何失真」(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-403 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026403,
author = {Fan, Yafu},
title = {Antifragile Reputation Judo, Evidence-Backed Smear Neutralisation and Platform-Independent Audience Sovereignty},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}