关于「钩子的作用模型:窗口、余波与留下来的是谁」,《直播间留存的生存分析》给出的定量结论是什么?
结论:第五章建立钩子的作用模型:窗口盖住尖峰时收益 693.4 观众分钟,领完就走时只剩 40.4;钩子留下的人此后的停留只有原本会留下者的 0.68。
数据口径:书中的直播间、观众数、风险参数、钩子效果与节目单均为示意;全部计算由书中程序生成;本书是非个人化经营建议,钩子只能是真实的内容或写明真实库存与真实截止时刻的福利,不用会重置的倒计时,不诱导未成年人消费,平台规则以官方公示为准。 第一章说明截尾让平均停留偏低:示意直播间里把下播当离开得 408.1 秒,只算已离开者得 278.0 秒,乘积极限估计的限定平均停留是 337.9 秒。
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: 40.4, 693.4
| no hook: 15493.7 audience-minutes, 29.6 of 1000 left at the end | ||||
| hook start | gain in audience-minutes | (switch at 900 s) | |||
| (s) | | | pi=0 | pi=0.5 | pi=1 |
| 780 | | | 141.6 | 74.2 | 7.5 |
| 810 | | | 136.6 | 71.6 | 7.3 |
| 840 | | | 131.7 | 69.0 | 7.0 |
| 870 | | | 693.4 | 349.2 | 24.0 |
| 900 | | | 688.9 | 355.5 | 40.4 |
| 930 | | | 118.8 | 61.9 | 5.6 |
| 960 | | | 115.0 | 59.9 | 5.4 |
| 990 | | | 111.4 | 58.1 | 5.3 |
| quiet 1500 | | 63.9 | 33.3 | 3.1 | |
| best start: pi=0 at 870 s | pi=0.5 at 900 s | pi=1 at 900 s |
查看完整验算程序
# Chapter 5. p51: where to put one hook around one switch spike.
# Exact audience curve of a closed group of 1000 settled viewers on the
# programme clock (1-second grid). Hazard per second: the two levels of
# M1 in p41 (rest 0.00072, switch windows 0.0069 for 30 s at each run-
# sheet switch). A hook is a real, pre-announced segment: for W = 60 s
# from its start the hazard is multiplied by 1 - rho (rho = 0.5, an
# illustrative effect, not a measured one); for E = 30 s after it ends,
# extra hazard returns pi * (hazard removed in the window), spread
# evenly: pi = 0 no exodus, pi = 1 everyone kept walks out right after.
import math
T_END = 3600
SWITCH = [180, 900, 1260, 1980, 2700] # 环节切换秒数(示意)
SPIKE_LEN = 30
REST, SPIKE = 0.00072, 0.0069 # p41 的两档估计(示意数据)
W, E, RHO = 60, 30, 0.5 # 钩子窗口、余波、风险降幅(示意)
def base_hazard():
h = [REST] * T_END
for s in SWITCH:
for t in range(s, s + SPIKE_LEN):
h[t] = SPIKE
return h
def with_hooks(h, starts, pi):
g = list(h)
for a in starts:
removed = math.fsum(RHO * h[t] for t in range(a, min(a + W,
T_END)))
for t in range(a, min(a + W, T_END)):
g[t] = h[t] * (1.0 - RHO)
for t in range(a + W, min(a + W + E, T_END)):
g[t] = g[t] + pi * removed / E
return g
def audience_minutes(g, n0=1000.0):
# 每秒风险常数:该秒内在场人秒 = N (1 - exp(-g)) / g
n, parts = n0, []
for x in g:
parts.append(n * (1.0 - math.exp(-x)) / x)
n *= math.exp(-x)
return math.fsum(parts) / 60.0, n
h = base_hazard()
base, end_n = audience_minutes(h)
print("no hook: %.1f audience-minutes, %.1f of 1000 left at the end"
% (base, end_n))
print("hook start | gain in audience-minutes (switch at 900 s)")
print(" (s) | pi=0 pi=0.5 pi=1")
best = {}
for a in range(780, 991, 30):
row = []
for pi in (0.0, 0.5, 1.0):
gain = audience_minutes(with_hooks(h, [a], pi))[0] - base
row.append(gain)
if pi not in best or gain > best[pi][0]:
best[pi] = (gain, a)
print("%8d | %6.1f %6.1f %6.1f" % (a, row[0], row[1], row[2]))
quiet = [audience_minutes(with_hooks(h, [1500], pi))[0] - base
for pi in (0.0, 0.5, 1.0)]
print("quiet 1500 | %6.1f %6.1f %6.1f" % tuple(quiet))
print("best start: " + " ".join(
"pi=%g at %d s" % (pi, best[pi][1]) for pi in (0.0, 0.5, 1.0)))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《直播间留存卡普兰迈耶生存曲线、划走风险尖峰定位与福利钩子插入秒数母典》第 5 章「钩子的作用模型:窗口、余波与留下来的是谁」(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-277 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026277,
author = {Fan, Yafu},
title = {Kaplan-Meier Retention Curves for Livestream Rooms: Locating Swipe-Away Hazard Spikes and Timing the Insertion of Hooks},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}