关于「母典页:内链拓扑如何把权威集中到一处」,《知识库权威传播与直答片段》给出的定量结论是什么?

结论:第二章证明星形内链下母典页的份额有一个与库的大小无关的上限 \(d/(1+d)\),即 0.4595。

数据口径:书中参数为示意;全部计算由书中程序生成;本书不是法律、理财或经营方面的个人化建议,也不对任何系统中的引用排名作出承诺。 第一章用幂迭代与随机冲浪者计算 PageRank 与 HITS:在阻尼系数 0.85 的示意知识库里,母典页拿到 0.3240,是均匀份额的 2.59 倍。

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: 0.4595

topology pages PR(m) x uniform Gini
ring, one link to m12 0.0345 0.41 0.097
star12 0.4662 5.59 0.383
star + ring12 0.2305 2.77 0.147
m > 3 sections > leaves15 0.2391 3.59 0.392
star closed form (1+dn)/(N(1+d))0.4662
star with11 articles: PR(m) = 0.4662, x uniform 5.6
star with50 articles: PR(m) = 0.4610, x uniform 23.5
star with200 articles: PR(m) = 0.4599, x uniform 92.4
star with 1000 articles: PR(m)0.4595, x uniform 460.0
limit d/(1+d)0.4595
查看完整验算程序
# Chapter 2. p21: how internal-link topology concentrates PageRank on a
# "mother page" (the canonical hub page of a knowledge base).
# Twelve pages: one mother page m and eleven articles. Four illustrative
# topologies, d = 0.85 (Brin and Page 1998). For the pure star (m links
# to every article, every article links only back to m) the closed form
# is PR(m) = (1 + d n) / (N (1 + d)), n = 11 articles, N = 12 pages,
# which tends to d / (1 + d) as n grows.
import math

D = 0.85
N_ART = 11


def pagerank(links, rounds=1000):
    pages = sorted(links)
    n = len(pages)
    pr = {p: 1.0 / n for p in pages}
    into = {p: [q for q in pages if p in links[q]] for p in pages}
    for _ in range(rounds):
        pr = {p: (1 - D) / n + D * math.fsum(
            pr[q] / len(links[q]) for q in into[p]) for p in pages}
    return pr


def gini(xs):
    xs = sorted(xs)
    n = len(xs)
    num = math.fsum((2 * (i + 1) - n - 1) * x for i, x in enumerate(xs))
    return num / (n * math.fsum(xs))


def art(i):
    return "a{:02d}".format(i % N_ART)


def topologies():
    ring = {art(i): [art(i - 1), art(i + 1)] for i in range(N_ART)}
    ring["m"] = [art(i) for i in range(N_ART)]
    ring[art(0)] = ring[art(0)] + ["m"]
    star = {art(i): ["m"] for i in range(N_ART)}
    star["m"] = [art(i) for i in range(N_ART)]
    both = {art(i): ["m", art(i - 1), art(i + 1)] for i in range(N_ART)}
    both["m"] = [art(i) for i in range(N_ART)]
    tree = {"m": ["s0", "s1", "s2"]}
    for s in range(3):
        kids = [art(i) for i in range(N_ART) if i % 3 == s]
        tree["s{}".format(s)] = kids + ["m"]
        for k in kids:
            tree[k] = ["s{}".format(s), "m"]
    return [("ring, one link to m", ring), ("star", star),
            ("star + ring", both), ("m > 3 sections > leaves", tree)]


if __name__ == "__main__":
    print("topology                  pages  PR(m)   x uniform   Gini")
    for name, links in topologies():
        pr = pagerank(links)
        n = len(pr)
        print("{:<26}{:4d}  {:.4f}  {:9.2f}  {:6.3f}".format(
            name, n, pr["m"], pr["m"] * n, gini(list(pr.values()))))
    closed = (1 + D * N_ART) / ((N_ART + 1) * (1 + D))
    print("star closed form (1+dn)/(N(1+d)) = {:.4f}".format(closed))
    for n in (11, 50, 200, 1000):
        h = (1 + D * n) / ((n + 1) * (1 + D))
        print("star with {:4d} articles: PR(m) = {:.4f}, x uniform {:.1f}"
              .format(n, h, h * (n + 1)))
    print("limit d/(1+d) = {:.4f}".format(D / (1 + D)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《机器可读知识库超链接权威传播、生成式问答检索截断博弈与高信噪比摘要提取》第 2 章「母典页:内链拓扑如何把权威集中到一处」(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-176 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026176,
  author    = {Fan, Yafu},
  title     = {Hyperlink Authority Propagation in Machine-Readable Knowledge Bases, Retrieval-Truncation Games in Generative Answering and High-SNR Snippet Extraction},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论