关于「七个关键词槽:合规过滤与背包选词」,《蓝海词与主题合集》给出的定量结论是什么?

结论:衰减系数的扫描给出同一个方向。系数取 0.3,表示同意图短语重复得很厉害,精确最优比天真做法高出 1.067 倍;系数取 0.9,同意图短语几乎各管各的,优势缩到 1.027 倍。重叠越重,「按分数取前七」越吃亏。

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: 0.3, 0.9, 1.027, 1.067

sets searched792
exact value 8.850, intents 5
information goods bundling2.94 bundle
left digit price effect2.43 psych
zipf law1.43 demand
bundle pricing math1.10 bundle
jaccard similarity0.70 cluster
screening menu design0.60 psych
chargeback ratio0.55 ops
greedy value 8.850
naive value 8.422, intents 4
information goods bundling2.94 bundle
left digit price effect2.43 psych
zipf law1.43 demand
bundle pricing math1.10 bundle
mixed bundling0.90 bundle
long tail sales0.80 demand
jaccard similarity0.70 cluster
exact over naive1.051
decay exact greedy naive exact/naive
0.38.450 8.450 7.920 1.067
0.68.850 8.850 8.422 1.051
0.99.250 9.250 9.005 1.027
查看完整验算程序
# Chapter 3. p32: choosing seven KDP keyword slots as a small knapsack.
# Input: the twelve phrases kept by p31, each with an illustrative
# expected weekly sales E (scored as in p22) and a search-intent group.
# Set value: phrases are taken in descending E; each phrase adds
#   E * (share of its words not yet in title or earlier phrases)
#     * DECAY^(number of earlier phrases in the same intent)
# because overlapping words and same-intent phrases reach the same
# shoppers. Exact optimum over all C(12, 7) sets versus a greedy pass
# and the naive "top seven by E". All data illustrative.
import itertools
import math

DECAY = 0.6
TITLE = set("niche keyword positioning thematic bundle direct sales".split())
PH = [
    ("information goods bundling", 2.94, "bundle"),
    ("bundle pricing math", 1.10, "bundle"),
    ("left digit price effect", 2.43, "psych"),
    ("zipf law", 1.43, "demand"),
    ("price discrimination", 0.30, "psych"),
    ("long tail sales", 0.80, "demand"),
    ("mixed bundling", 0.90, "bundle"),
    ("screening menu design", 0.60, "psych"),
    ("refund rate model", 0.40, "ops"),
    ("chargeback ratio", 0.55, "ops"),
    ("jaccard similarity", 0.70, "cluster"),
    ("poisson cash flow", 0.35, "ops"),
]


def value(idx, decay=DECAY):
    seen = set(TITLE)
    per = {}
    parts = []
    for i in sorted(idx, key=lambda j: -PH[j][1]):
        words = PH[i][0].split()
        fresh = len([w for w in words if w not in seen]) / len(words)
        k = per.get(PH[i][2], 0)
        parts.append(PH[i][1] * fresh * decay ** k)
        per[PH[i][2]] = k + 1
        seen.update(words)
    return math.fsum(parts)


def greedy(k=7, decay=DECAY):
    pick = []
    for _ in range(k):
        best = max((j for j in range(len(PH)) if j not in pick),
                   key=lambda j: (value(pick + [j], decay), -j))
        pick.append(best)
    return pick


def show(name, idx):
    groups = sorted({PH[i][2] for i in idx})
    print("{:<8} value {:.3f}, intents {}".format(
        name, value(idx), len(groups)))
    for i in sorted(idx, key=lambda j: -PH[j][1]):
        print("    {:<28} {:.2f}  {}".format(PH[i][0], PH[i][1], PH[i][2]))


if __name__ == "__main__":
    n = len(PH)
    count = 0
    best, best_v = None, -1.0
    for c in itertools.combinations(range(n), 7):
        count += 1
        v = value(list(c))
        if v > best_v + 1e-12:
            best, best_v = list(c), v
    naive = sorted(range(n), key=lambda j: -PH[j][1])[:7]
    print("sets searched: {:,}".format(count))
    show("exact", best)
    g = greedy()
    print("greedy   value {:.3f}".format(value(g)))
    show("naive", naive)
    print("exact over naive: {:.3f}".format(best_v / value(naive)))
    print("decay   exact   greedy   naive   exact/naive")
    for d in (0.3, 0.6, 0.9):
        ex = max(value(list(c), d)
                 for c in itertools.combinations(range(n), 7))
        gr = value(greedy(7, d), d)
        nv = value(naive, d)
        print("  {:.1f}  {:>6.3f}  {:>7.3f}  {:>6.3f}  {:>11.3f}".format(
            d, ex, gr, nv, ex / nv))

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 3 章「七个关键词槽:合规过滤与背包选词」(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