关于「合规评论速度:邀请漏斗、泊松区间与「前三名」的模拟分位数」,《地图排名与同城承接》给出的定量结论是什么?

结论:第四章讲合规评论速度:全部客人一视同仁的邀请漏斗、精确泊松区间,以及把「挤进前三名」写成模拟分位数:每月 12 条评论时,一年内达标的概率是 0.843。

数据口径:本书把三因子写成乘积式示意打分,全部权重都是示意参数,不是谷歌真实权重:一家距离两公里、类目登记错误的示意门店,要靠评论追平改对主类目的效果,需要 8,511 条评论。 书中参数为示意,全部计算由书中程序生成。

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: 0.843, 12

goal: top 3 at >= 19 of 25 points; 49 means not by month 48
lam/monthP10P50P90P(<=12 months)
34949490.000
64949490.000
82226300.000
101316180.025
12911130.843
157791.000
204561.000
查看完整验算程序
# Chapter 4. p43: "top three on the map" as a simulated quantile, not a
# promise. Synthetic stores of p22. Each month the store gains
# Poisson(lam) reviews from inviting every customer; each rival gains
# Poisson(its own velocity). Ratings stay at their means. Goal: the
# store in the top 3 at >= 19 of 25 grid points (what +200 reviews gave
# in p22). 400 runs per velocity, capped at 48 months, seed fixed.
import math
import random

CAT = {"primary": 1.0, "secondary": 0.55}
RIVALS = [("primary", 0.8, (1.0, 0.5), 25, 4.4, 2),
          ("primary", 0.7, (-1.2, 0.8), 180, 4.5, 6),
          ("secondary", 0.9, (0.6, -1.5), 90, 4.6, 4),
          ("primary", 0.5, (-2.0, -2.0), 400, 4.3, 10),
          ("primary", 0.6, (2.2, 2.0), 60, 4.5, 3)]
OWN = ("secondary", 0.6, (0.0, 0.0), 40, 4.7, 3)
PTS = [(gx, gy) for gy in (2, 1, 0, -1, -2) for gx in (-2, -1, 0, 1, 2)]


def score(store, pt):
    cat, svc, (x, y), n, mean, v = store
    d = math.hypot(x - pt[0], y - pt[1])
    rbar = (10.0 * 4.3 + n * mean) / (10.0 + n)
    prom = (1.0 + 0.30 * math.log1p(n)) * (rbar / 5.0) ** 2
    prom *= 1.0 + 0.15 * (1.0 - math.exp(-v / 8.0))
    return CAT[cat] * (0.7 + 0.3 * svc) * math.exp(-d / 1.5) * prom


def top3(own, rivals):
    c = 0
    for pt in PTS:
        s = score(own, pt)
        r = 1
        for riv in rivals:
            if score(riv, pt) > s:
                r += 1
        if r <= 3:
            c += 1
    return c


def poisson(rng, mu):
    lim, k, p = math.exp(-mu), 0, 1.0
    while True:
        p *= rng.random()
        if p < lim:
            return k
        k += 1


def run(rng, lam, cap=48):
    own = list(OWN)
    rivals = [list(r) for r in RIVALS]
    own[5] = lam
    for month in range(1, cap + 1):
        own[3] += poisson(rng, lam)
        for r in rivals:
            r[3] += poisson(rng, r[5])
        if top3(tuple(own), [tuple(r) for r in rivals]) >= 19:
            return month
    return cap + 1


def q(xs, p):
    s = sorted(xs)
    return s[min(len(s) - 1, int(p * len(s)))]


if __name__ == "__main__":
    rng = random.Random(20261120)
    print("goal: top 3 at >= 19 of 25 points; 49 means not by month 48")
    print("lam/month   P10  P50  P90   P(<=12 months)")
    for lam in (3, 6, 8, 10, 12, 15, 20):
        res = [run(rng, lam) for _ in range(400)]
        hit = 0
        for m in res:
            if m <= 12:
                hit += 1
        print("{:>9}   {:>3}  {:>3}  {:>3}   {:.3f}".format(
            lam, q(res, 0.1), q(res, 0.5), q(res, 0.9), hit / 400.0))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《本地搜索地图排名因子、评论速度信号权重与同城高急迫线索截流母典》第 4 章「合规评论速度:邀请漏斗、泊松区间与「前三名」的模拟分位数」(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-369 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026369,
  author    = {Fan, Yafu},
  title     = {Local-Pack Search Ranking Factors, Review-Velocity Signalling and High-Urgency Geo-Intent Lead Capture},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论