关于「推流冒充口碑:非平稳外生率与伪临界」,《评论裂变分支比》给出的定量结论是什么?

结论:第六章说明推流冒充口碑的伪临界:真实分支比 0.3 的推流帖被朴素估计成 0.864,把推流窗口写进模型后为 0.276,朴素估计把推流产出高估 5.15 倍。

数据口径:书中的帖子、账号、流量、成本与评论价值均为示意参数;全部计算由书中程序生成;本书是非个人化经营建议,只讨论真实用户的评论与平台官方工具买来的曝光,不涉及任何制造评论或虚构稀缺的手段。 第四章用小崎递推、补偿子恒等式与黄金分割求最大似然估计,示意帖的分支比 95 % 区间从 0.42 到 0.92,两小时估计的标准差约 0.186。

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: 0.276, 0.3, 0.864

poststrue ncommentsnaive nwindow npush ratio r
boosted0.3216.00.8640.2766.65
organic0.6129.60.4870.4390.28
true push ratio for boosted posts: r = 6.0
查看完整验算程序
# Chapter 6. p61: Paid reach mistaken for word of mouth.
# Posts: background outside rate 0.5 per minute, beta = 0.1, T = 120.
# "Boosted" posts get extra outside traffic of 3.0 per minute from
# minute 30 to minute 60 (a paid push) and have a true n of 0.3.
# "Organic" posts get no push and have n = 0.6. Two fits, beta fixed:
#   naive   constant mu (p42 estimator)
#   window  mu0 (1 + r 1[30 <= t < 60]): the push window, which the
#           operator knows from the ad log, enters as a covariate;
#           mu0 from the compensator identity, r and n by golden search.
# Filimonov and Sornette (2015) show that ignored regime shifts bias n
# upward. Means of 30 posts per row; fixed seed; illustrative.
import math
import random

BETA, T, BASE, PUSH, A, B = 0.1, 120.0, 0.5, 3.0, 30.0, 60.0
G = (math.sqrt(5) - 1) / 2


def simulate(n, boosted, rng):
    t, excess, times = 0.0, 0.0, []
    top = BASE + (PUSH if boosted else 0.0)
    while True:
        bound = top + excess
        w = rng.expovariate(bound)
        t += w
        if t > T:
            return times
        excess *= math.exp(-BETA * w)
        mu = BASE + (PUSH if boosted and A <= t < B else 0.0)
        if rng.random() * bound <= mu + excess:
            times.append(t)
            excess += n * BETA


def stats(times, beta, horizon):
    a, prev, out = 0.0, None, []
    for t in times:
        if prev is not None:
            a = math.exp(-beta * (t - prev)) * (1 + a)
        out.append(a)
        prev = t
    s = math.fsum(1 - math.exp(-beta * (horizon - t)) for t in times)
    return out, s


def loglik(mu, n, beta, a_list, s, horizon):
    terms = [math.log(mu + n * beta * a) for a in a_list]
    return math.fsum(terms) - mu * horizon - n * s


def golden_max(f, lo, hi, iters=40):
    x1, x2 = hi - G * (hi - lo), lo + G * (hi - lo)
    f1, f2 = f(x1), f(x2)
    for _ in range(iters):
        if f1 < f2:
            lo, x1, f1 = x1, x2, f2
            x2 = lo + G * (hi - lo)
            f2 = f(x2)
        else:
            hi, x2, f2 = x2, x1, f1
            x1 = hi - G * (hi - lo)
            f1 = f(x1)
    x = (lo + hi) / 2
    return x, f(x)


def fit_n(times, beta, horizon):
    a_list, s = stats(times, beta, horizon)
    cnt = len(times)
    hi = min(0.999, cnt / s - 1e-9)
    n, ll = golden_max(lambda v: loglik((cnt - v * s) / horizon, v, beta,
                                        a_list, s, horizon), 0.0, hi)
    return n, (cnt - n * s) / horizon, ll



def fit_window(times):
    a_list, s = stats(times, BETA, T)
    flag = [1.0 if A <= t < B else 0.0 for t in times]
    cnt = len(times)

    def ll_r(r):
        expo = T + r * (B - A)

        def ll_n(n):
            mu0 = (cnt - n * s) / expo
            terms = [math.log(mu0 * (1 + r * f) + n * BETA * a)
                     for a, f in zip(a_list, flag)]
            return math.fsum(terms) - mu0 * expo - n * s
        return golden_max(ll_n, 0.0, min(0.999, cnt / s - 1e-9), 30)

    r, _ = golden_max(lambda x: ll_r(x)[1], 0.0, 20.0, 30)
    return ll_r(r)[0], r


rng = random.Random(20261113)
print("posts     true n  comments  naive n  window n  push ratio r")
for label, n, boosted in (("boosted", 0.3, True), ("organic", 0.6, False)):
    cnt, nv, wn, rs = [], [], [], []
    for _ in range(30):
        ts = simulate(n, boosted, rng)
        cnt.append(len(ts))
        nv.append(fit_n(ts, BETA, T)[0])
        w, r = fit_window(ts)
        wn.append(w)
        rs.append(r)
    print("%-8s  %.1f     %6.1f    %.3f    %.3f     %.2f"
          % (label, n, math.fsum(cnt) / 30, math.fsum(nv) / 30,
             math.fsum(wn) / 30, math.fsum(rs) / 30))
print("true push ratio for boosted posts: r = %.1f" % (PUSH / BASE))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《霍克斯自激点过程、评论裂变分支比临界值与直播跟风抢购余震半衰期母典》第 6 章「推流冒充口碑:非平稳外生率与伪临界」(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-234 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026234,
  author    = {Fan, Yafu},
  title     = {Hawkes Self-Exciting Point Processes: Branching-Ratio Criticality in Comment Cascades and the Aftershock Half-Life of Livestream Herd Buying},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论