关于「J 型从哪里来:U 形发帖意愿与两种自选择偏差」,《星级自选择与贝叶斯排序》给出的定量结论是什么?
结论:第一章用「购买偏差加少报偏差」的两段筛子解释评论星级的 J 型分布,示意参数下购买偏差使均值上移 +0.532 星。
数据口径:书中的顾客、商品与参数均为示意模型;全部计算由书中程序生成;本书是非个人化的经营参考,不是针对具体店铺的经营建议,也不是法律意见,书中提到的法规以上架前人工复核为准;本书反对一切刷单、购买评论与筛选式求评。 第一章用「购买偏差加少报偏差」的两段筛子解释评论星级的 J 型分布,示意参数下购买偏差使均值上移 +0.532 星。
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: 0.532
| star shares: public, buyers, reviewers (illustrative) | |
| group s | 1 s=2 s=3 s=4 s=5 mean |
| public | 0.087 0.146 0.334 0.305 0.128 3.240 |
| buyers | 0.061 0.034 0.232 0.418 0.255 3.772 |
| reviewers | 0.171 0.032 0.087 0.234 0.476 3.812 |
| acquisition bias (buyers - public): +0.532 | |
| under-reporting (reviewers - buyers): +0.040 | |
| total gap (reviewers - public): +0.572 | |
| buy threshold scan: c, buyer mean, reviewer mean | |
| c | -0.5 3.593 3.577 |
| c=+0.3 | 3.772 3.812 |
| c=+1.0 | 3.945 4.015 |
查看完整验算程序
# Chapter 1. p12: split the gap between reviews and the public in two biases.
# Model (after the two self-selection biases named by Hu, Pavlou and Zhang):
# latent fit f ~ N(0, 1) in the whole public;
# satisfaction x = 3.4 + 1.0 f, cut at 1.5/2.5/3.5/4.5 into 1..5 stars;
# acquisition: a shopper buys if f + e > c, e ~ N(0, 1.2), c = 0.3;
# a unit is faulty with prob 0.06 whatever the fit, and then gets 1 star;
# under-reporting: a buyer with star s posts with U-shaped prob w(s).
# Integrals by a fine midpoint grid on f; every parameter is illustrative.
import math
CUTS = (1.5, 2.5, 3.5, 4.5)
W = (0.30, 0.10, 0.04, 0.06, 0.20)
SD_E, C = 1.2, 0.3
FAULT = 0.06
def phi(z):
return math.exp(-0.5 * z * z) / math.sqrt(2.0 * math.pi)
def Phi(z):
return 0.5 * (1.0 + math.erf(z / math.sqrt(2.0)))
def star(x):
s = 1
for c in CUTS:
if x > c:
s += 1
return s
def dists(c=C):
pop = [[] for _ in range(5)]
buy = [[] for _ in range(5)]
rev = [[] for _ in range(5)]
h = 0.002
n = int(16.0 / h)
for i in range(n):
f = -8.0 + (i + 0.5) * h
d = phi(f) * h
s = star(3.4 + f) - 1
pb = Phi((f - c) / SD_E)
for k, q in ((s, 1.0 - FAULT), (0, FAULT)):
pop[k].append(d * q)
buy[k].append(d * pb * q)
rev[k].append(d * pb * q * W[k])
out = []
for g in (pop, buy, rev):
m = [math.fsum(v) for v in g]
tot = math.fsum(m)
out.append([x / tot for x in m])
return out
def mean(sh):
return math.fsum((k + 1) * s for k, s in enumerate(sh))
pop, buy, rev = dists()
print("star shares: public, buyers, reviewers (illustrative)")
print("%-10s %6s %6s %6s %6s %6s %6s" % (
"group", "s=1", "s=2", "s=3", "s=4", "s=5", "mean"))
for name, sh in (("public", pop), ("buyers", buy), ("reviewers", rev)):
print("%-10s %s %6.3f" % (
name, " ".join("%6.3f" % s for s in sh), mean(sh)))
a = mean(buy) - mean(pop)
u = mean(rev) - mean(buy)
print("acquisition bias (buyers - public): %+.3f" % a)
print("under-reporting (reviewers - buyers): %+.3f" % u)
print("total gap (reviewers - public): %+.3f" % (a + u))
print("buy threshold scan: c, buyer mean, reviewer mean")
for c in (-0.5, 0.3, 1.0):
_, b2, r2 = dists(c)
print(" c=%+.1f %.3f %.3f" % (c, mean(b2), mean(r2)))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《评论星级J型自选择偏差、贝叶斯平均排序与九品中正评级通胀母典》第 1 章「J 型从哪里来:U 形发帖意愿与两种自选择偏差」(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-212 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026212,
author = {Fan, Yafu},
title = {J-Shaped Self-Selection in Star Ratings, Bayesian-Average Ranking and Rating Inflation from the Nine-Rank System to Online Marketplaces},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}