关于「口语化压缩:双人播客的广度与重复、换手时长与香农命名」,《专著改编的信息保留代数》给出的定量结论是什么?
结论:第二章显示三十分钟的播客只讲约 10 个论点时保留 0.4820,把 20 个论点各讲一遍只保留 0.3117,单人连续讲述的最优时长约 41.5 秒。
数据口径:书中参数为示意;全部计算由书中程序生成;本书不是法律意见,也不是个人化的经营或投资建议。 第六章在 40 个制作工时下求出播客、投影片与题库的最优组合,价值 2.0685,不做题库只有 1.4164;事实核对应放在上游;每周 2 次提醒使期望学习次数最高,为 6.2792,过量提醒有害,对脆弱学习者阈值更低;只改编有权改编的书,合成内容必须标识。
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: 0.3117, 0.4820, 10, 20
| ideas | share of book | retained weight | exposures (top, last) | |
| 2 | 0.0500 | 0.2604 | 20 | 18 |
| 4 | 0.1000 | 0.3801 | 10 | 8 |
| 6 | 0.1500 | 0.4457 | 7 | 5 |
| 8 | 0.2000 | 0.4752 | 6 | 3 |
| 10 | 0.2500 | 0.4820 | 5 | 2 |
| 12 | 0.3000 | 0.4763 | 5 | 1 |
| 14 | 0.3500 | 0.4624 | 5 | 1 |
| 16 | 0.4000 | 0.4408 | 4 | 1 |
| 18 | 0.4500 | 0.4022 | 3 | 1 |
| 20 | 0.5000 | 0.3117 | 1 | 1 |
| best breadth 10 ideas, retained weight 0.4820 | ||||
| stating all 40 once would need 60.0 minutes |
查看完整验算程序
# Chapter 2. p21: breadth against repetition in a 30-minute two-host show.
# The monograph holds 40 ideas with importance w_i proportional to
# 1 / i^0.8. A listener cannot rewind, so an idea stated once is kept with
# probability q = 0.40; each restatement (an example, a question from the
# second host, a recap) is another independent chance, so k exposures
# keep it with 1 - (1 - q)^k. Stating an idea costs 1.5 minutes, each
# restatement 0.75. For every breadth n (top n ideas) the spare minutes
# go greedily to the restatement with the largest weighted gain; the
# objective is separable and concave in k, so greedy is exact here.
# Times, q and the importance curve illustrative.
import math
N_IDEAS, MINUTES, FIRST, AGAIN, Q = 40, 30.0, 1.5, 0.75, 0.40
W = [1.0 / (i + 1) ** 0.8 for i in range(N_IDEAS)]
TOTAL = math.fsum(W)
def keep(k):
return 1.0 - (1.0 - Q) ** k
def best_for(n):
spare = MINUTES - n * FIRST
if spare < 0:
return None
k = [1] * n
for _ in range(int(spare / AGAIN + 1e-9)):
j = max(range(n), key=lambda i: W[i] * (keep(k[i] + 1)
- keep(k[i])))
k[j] += 1
kept = math.fsum(W[i] * keep(k[i]) for i in range(n))
return kept / TOTAL, k
print("ideas share of book retained weight exposures (top, last)")
best = (0.0, 0)
for n in (2, 4, 6, 8, 10, 12, 14, 16, 18, 20):
r = best_for(n)
if r is None:
continue
frac, k = r
best = max(best, (frac, n))
print("{:5d} {:13.4f} {:15.4f} {:9d} {:9d}".format(
n, n / N_IDEAS, frac, k[0], k[-1]))
print("best breadth {} ideas, retained weight {:.4f}".format(
best[1], best[0]))
print("stating all 40 once would need {:.1f} minutes".format(
N_IDEAS * FIRST))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《一书多吃内容裂变代数:从硬核专著自动编译双人播客、导学投影片与通关题库》第 2 章「口语化压缩:双人播客的广度与重复、换手时长与香农命名」(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-151 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026151,
author = {Fan, Yafu},
title = {Content Fission Algebra: Compiling Hard-Core Monographs into Two-Host Podcasts, Study Slides and Quiz Banks},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}