关于「幂迭代与阻尼系数:超链接权威在知识库里怎样流动」,《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。

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: 0.3240, 0.85, 2.59

paged=0.50d=0.70d=0.85d=0.95
canon0.24540.29000.32400.3474
def-a0.11900.11610.11230.1087
def-b0.12970.13390.13710.1390
faq0.10200.09410.08750.0825
data0.11540.11820.12290.1273
source0.11180.11270.11690.1217
blog0.11420.09760.08060.0672
old0.06250.03750.01880.0063
iterations16202529
d-bound33.264.6141.7448.9
d=0.50: canon holds 0.2454, 1.96 times the uniform 1/8
d=0.70: canon holds 0.2900, 2.32 times the uniform 1/8
d=0.85: canon holds 0.3240, 2.59 times the uniform 1/8
d=0.95: canon holds 0.3474, 2.78 times the uniform 1/8
查看完整验算程序
# Chapter 1. p11: PageRank by power iteration on an illustrative
# eight-page knowledge base, for four damping factors d.
# Model: Brin and Page (1998), PR(p) = (1-d)/N + d * sum over pages q
# linking to p of PR(q)/out(q). The link graph is illustrative, not
# measured; d = 0.85 is the value Brin and Page report, the others are
# shown for contrast. Iteration stops when the L1 change is below 1e-10.
import math

PAGES = ["canon", "def-a", "def-b", "faq", "data", "source", "blog",
         "old"]
LINKS = {
    "canon": ["def-a", "def-b", "faq", "data", "source", "blog"],
    "def-a": ["canon", "def-b"],
    "def-b": ["canon", "data"],
    "faq": ["canon", "def-a", "def-b"],
    "data": ["canon", "source"],
    "source": ["canon"],
    "blog": ["canon", "faq", "def-a"],
    "old": ["blog"],
}


def pagerank(d, tol=1e-10, cap=10000):
    n = len(PAGES)
    pr = {p: 1.0 / n for p in PAGES}
    for it in range(1, cap + 1):
        new = {}
        for p in PAGES:
            inflow = math.fsum(pr[q] / len(LINKS[q]) for q in PAGES
                               if p in LINKS[q])
            new[p] = (1 - d) / n + d * inflow
        diff = math.fsum(abs(new[p] - pr[p]) for p in PAGES)
        pr = new
        if diff < tol:
            return pr, it
    return pr, cap


if __name__ == "__main__":
    ds = (0.50, 0.70, 0.85, 0.95)
    runs = [pagerank(d) for d in ds]
    print("page      " + "".join("  d={:.2f}".format(d) for d in ds))
    for p in PAGES:
        print("{:<10}".format(p)
              + "".join("  {:6.4f}".format(r[0][p]) for r in runs))
    print("iterations" + "".join("  {:6d}".format(r[1]) for r in runs))
    print("d-bound   " + "".join(
        "  {:6.1f}".format(math.log(1e-10) / math.log(d)) for d in ds))
    for d, r in zip(ds, runs):
        print("d={:.2f}: canon holds {:.4f}, {:.2f} times the uniform"
              " 1/8".format(d, r[0]["canon"], r[0]["canon"] * 8))

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 1 章「幂迭代与阻尼系数:超链接权威在知识库里怎样流动」(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