关于「进前位概率与蓝海词评分」,《蓝海词与主题合集》给出的定量结论是什么?

结论:数字说得很直白。大词上,新书评价为零时进前三的概率四舍五入是 0.0000,五条评价时是 0.0001,二十条评价也只有 0.0002。换算成更直观的说法:五条评价的新书,大约每 13,657 次搜索才有一次出现在前三位。中等词好一些,五条评价时是 0.0117。冷僻词完全是另一个世界:零评价时已有 0.1559,五条评价时是 0.3753,二十条评价时是 0.6015。同样五条评价,冷僻词上的机会是大词的 5,126 倍。

数据口径:对小作者,关键词带来的期望销量约等于搜索需求、进入前几位的概率、点击率与转化率之积;大词的进前位概率近乎为零,示意数据里周搜索两万次的大词只带来约 0.0254 本周销量,冷僻短语却有 3.1445 本。 核对原文后,本书做了蓝图修正:排名换销量的指数是示意参数而非定值;KDP 关键词规则以官方帮助页为准,每本书至多七个关键词,不得用与本书无关的作者名或自己不拥有的品牌;AI 生成内容须如实披露;本书不引用艾瑞里的任何研究。

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: 0.0000, 0.0001, 0.0002, 0.0117, 0.1559, 0.3753

P(new title in top 3), Plackett-Luce with Gumbel noise
classcompsreviews 0reviews 5reviews 20
head400.00000.00010.0002
mid150.00400.01170.0247
niche60.15590.37530.6015
head, 5 reviews: about 1 search page in 13,657
niche over head at 5 reviews: 5,126 times

核心公式

\[E = D \sum_{i=1}^{3} P(\text{第 } i \text{ 位}) \cdot R(1-R)^{i-1} \cdot \text{CVR}\]
查看完整验算程序
# Chapter 2. p21: chance that a new title reaches the top three results.
# Ranking model (illustrative): each listing has strength
#   u = BETA * ln(1 + reviews) + GAMMA * ln(1 + weekly sales) + rel
# and the results page orders listings by u plus Gumbel noise, which is
# a Plackett-Luce (multinomial logit) ranking. Competitors per keyword
# class are drawn once (fixed seed); the new title has few reviews and
# almost no sales history. rel = topical match of the new title.
# P(top 3) is integrated numerically over the Gumbel noise.
import math
import random

BETA, GAMMA = 0.6, 0.5
CLASSES = [
    # name, competitors, median reviews, median weekly sales, rel new
    ("head", 40, 3_000, 400, 0.0),
    ("mid", 15, 150, 20, 0.5),
    ("niche", 6, 8, 1, 1.0),
]


def competitors(rng, n, rev, sales):
    out = []
    for _ in range(n):
        r = rev * math.exp(rng.gauss(0, 1.0))
        s = sales * math.exp(rng.gauss(0, 1.0))
        out.append(BETA * math.log1p(r) + GAMMA * math.log1p(s))
    return out


def p_top(comp, u_new, k=3, step=0.002):
    # exact up to quadrature: integrate over the new title's noisy score
    # x = u_new + G, Gumbel density, and the Poisson-binomial chance that
    # fewer than k competitors beat x
    terms = []
    for i in range(int(round(46.0 / step))):
        x = u_new - 6.0 + i * step
        z = x - u_new
        dens = math.exp(-z - math.exp(-z))
        dp = [1.0] + [0.0] * k
        for u in comp:
            q = 1.0 - math.exp(-math.exp(u - x))
            dp = [dp[0] * (1 - q)] + [dp[j] * (1 - q) + dp[j - 1] * q
                                      for j in range(1, k + 1)]
        terms.append(dens * math.fsum(dp[:k]) * step)
    return math.fsum(terms)


if __name__ == "__main__":
    rng = random.Random(20261120)
    pool = {c[0]: competitors(rng, c[1], c[2], c[3]) for c in CLASSES}
    print("P(new title in top 3), Plackett-Luce with Gumbel noise")
    print("  class   comps   reviews 0   reviews 5   reviews 20")
    table = {}
    for name, n, _, _, rel in CLASSES:
        row = []
        for rv in (0, 5, 20):
            u = BETA * math.log1p(rv) + GAMMA * math.log1p(0.2) + rel
            row.append(p_top(pool[name], u))
        table[name] = row
        print("  {:<6} {:>6}   {:>9.4f}   {:>9.4f}   {:>10.4f}".format(
            name, n, *row))
    print("head, 5 reviews: about 1 search page in {:,.0f}".format(
        1 / table["head"][1]))
    print("niche over head at 5 reviews: {:,.0f} times".format(
        table["niche"][1] / table["head"][1]))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《百卷数字专著蓝海词卡位、高客单主题合集直售与全自动知识收银母典》第 2 章「进前位概率与蓝海词评分」(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-356 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026356,
  author    = {Fan, Yafu},
  title     = {Niche Keyword Positioning for Digital Monograph Catalogues, High-Ticket Thematic Bundle Direct Sales and Automated Knowledge Cashflow},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论