关于「评论区从众:阿希实验、信息级联与累积优势」,《直播开场、评论区与收口的计算》给出的定量结论是什么?

结论:第四章用精确马尔可夫链求解信息级联,一条假推荐就把错误级联的概率放大到 2.63 倍。

数据口径:书中参数均为示意;全部计算由书中程序生成;本书不构成个别化的法律、广告合规或营销建议。 第一章用伽马-泊松模型计算开场七个音节的贝叶斯惊奇度,留存与惊奇度呈倒 U 关系,示意模型的最优值为 1.79 比特,九成增益区间为 1.13 到 2.62 比特,「七个字」取决于语速。

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: 2.63

true quality low; P(right cascade) / P(wrong cascade)
qno seed1 fake seed2 fake seeds
0.550.599 / 0.4010.329 / 0.6710.000 / 1.000
0.600.692 / 0.3080.415 / 0.5850.000 / 1.000
0.700.845 / 0.1550.591 / 0.4090.000 / 1.000
0.800.941 / 0.0590.753 / 0.2470.000 / 1.000
0.900.988 / 0.0120.889 / 0.1110.000 / 1.000
q 0.70: one fake seed multiplies the wrong-cascade risk by 2.63
查看完整验算程序
# Chapter 4. p41: Information cascades in a comment section, exactly.
# Model of Bikhchandani, Hirshleifer and Welch: viewers decide in turn
# whether to endorse a product whose true quality is low. Each has a
# private signal that is right with probability q, and sees all earlier
# public actions. With the tie rule "follow your own signal", the
# difference d = endorsements - rejections moves by one step until
# |d| = 2; from then on everyone copies (a cascade). The chain is solved
# exactly with fractions. "Seeds" are fake endorsements placed before
# the first real viewer: d starts at 1 or 2. The q values are
# illustrative; the model is the published one.
from fractions import Fraction as F


def absorb(q, start, n=60):
    # probability mass on d in {-1, 0, 1}; absorbing at -2 (right), +2
    dist = {start: F(1)} if abs(start) < 2 else {}
    right = F(0)
    wrong = F(1) if start >= 2 else F(0)
    for _ in range(n):
        new = {}
        for d, p in dist.items():
            for step, w in ((-1, q), (1, 1 - q)):
                e = d + step
                if e <= -2:
                    right += p * w
                elif e >= 2:
                    wrong += p * w
                else:
                    new[e] = new.get(e, F(0)) + p * w
        dist = new
    return float(right), float(wrong)


print("true quality low; P(right cascade) / P(wrong cascade)")
print("  q     no seed          1 fake seed      2 fake seeds")
for q in (F(11, 20), F(3, 5), F(7, 10), F(4, 5), F(9, 10)):
    cells = []
    for s in (0, 1, 2):
        r, w = absorb(q, s)
        cells.append("{:.3f} / {:.3f}".format(r, w))
    print("  {:.2f}  {}".format(float(q), "   ".join(cells)))
r0, w0 = absorb(F(7, 10), 0)
r1, w1 = absorb(F(7, 10), 1)
print("q 0.70: one fake seed multiplies the wrong-cascade risk by {:.2f}"
      .format(w1 / w0))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《短视频与直播前七字听觉劫持、评论区从众引爆与脉冲逼单剧本学》第 4 章「评论区从众:阿希实验、信息级联与累积优势」(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-147 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026147,
  author    = {Fan, Yafu},
  title     = {Seven-Syllable Auditory Hooks in Short Video and Livestreams, Comment-Section Herding Ignition and Pulse-Closing Scripts},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论