关于「可见度的度量:关键词堆砌为何失灵,出处、数字与引语为何有效」,《生成式搜索引用权重》给出的定量结论是什么?

结论:第一章还原原文效应表,并用比例分配模型说明单独采用的改写只换来 0.3027 的份额提升、全部采用时归零。

数据口径:书中参数均为示意;全部计算由书中程序生成。 第五章讲可核查性,并用模拟说明从外部推断特征权重的脆弱:漏掉长度时统计数字的权重被高估 0.3252。

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: 0.3027

sources per answer N = 5, weight multiplier g = 1.4093
adoptersadopter shrgain vs 1/Nnon-adopter
0--0.2000
10.2605+0.30270.1849
20.2422+0.21100.1719
30.2263+0.13140.1606
40.2123+0.06170.1507
50.2000+0.0000-
lone adopter: weight +0.4093 buys share +0.3027
all adopt: every share back to 0.2000, gain 0
查看完整验算程序
# Chapter 1. p13: why a rewrite gain shrinks when competitors adopt it.
# Model: an answer splits a fixed attention budget among N sources in
# proportion to weights. A rewrite multiplies a source's weight by g.
# With a adopters among N, an adopter's share is g / (a g + N - a);
# its gain over the plain share 1 / N falls to zero when all adopt.
# g = 1.4093 is the best PAWC ratio of p11 (quotation addition); the
# proportional-share model is ours, an illustration of the congestion
# reported by Puerto et al. (2025), not their data.
from fractions import Fraction as F

N = 5
G = F(272, 193)   # 27.2 / 19.3 from Aggarwal et al. Table 1

print("sources per answer N = %d, weight multiplier g = %.4f" % (
    N, float(G)))
print("%9s %13s %13s %13s" % ("adopters", "adopter shr", "gain vs 1/N",
                             "non-adopter"))
for a in range(0, N + 1):
    total = a * G + (N - a)
    ad = G / total if a else None
    non = 1 / total if a < N else None
    print("%9d %13s %13s %13s" % (
        a, "-" if ad is None else "%.4f" % float(ad),
        "-" if ad is None else "%+.4f" % float(ad * N - 1),
        "-" if non is None else "%.4f" % float(non)))
lone = G / (G + N - 1) * N - 1
print("lone adopter: weight +%.4f buys share +%.4f" % (
    float(G - 1), float(lone)))
print("all adopt: every share back to %.4f, gain 0" % (1 / N))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《生成式搜索引用权重几何学、问答引擎检索重排特征与高被引双语直答页段落拓扑母典》第 1 章「可见度的度量:关键词堆砌为何失灵,出处、数字与引语为何有效」(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-260 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026260,
  author    = {Fan, Yafu},
  title     = {The Geometry of Citation Weight in Generative Search: Retrieval and Reranking Features and the Paragraph Topology of Highly Cited Bilingual Answer Pages},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论