5D Library › 短视频完播率风险函数、推荐反馈环与自动化获客漏斗 · English

第七章 揣情、章句与买椟还珠

为什么在这里读这些文献

本书的三个问题,即开头如何决定观众是否留下、何时向人提出请求、包装与内容的关系,在中国古典文献中都有经典的论述。本章读三段文字:《鬼谷子·揣篇》论揣度人情,《文心雕龙·章句》论开篇与全篇的关系,《韩非子·外储说左上》的「买椟还珠」。引文均已与所列底本逐字核对,标点依底本。

推导

# Chapter 7. p14: Three classical sayings, three numbers from this book.
# 1. Wenxin Diaolong, "Zhangju": the opening words anticipate the
#    middle. With the Weibull swipe model of p01 (k = 0.6, L = 20 s),
#    what share of all swipe-aways in a 60 s video happen in the first
#    3 s and the first 5 s?
# 2. Guiguzi, "Chuai": go to a person at the height of his delight.
#    With the attention residue model of p08 (rho = 0.5, tau = 8 s,
#    beta = 0.03 per second), how much better is the offer at the best
#    moment than at the instant of arrival?
# 3. Hanfeizi, the man of Zheng who bought the casket and returned the
#    pearl. With the hook breadth model of p07, what share of viewers
#    attracted by a hook of breadth b are "casket buyers", people the
#    hook brought in who would never buy, judged against the narrowest
#    hook: 1 - 20^(-b^2)?
import math

def surv(t, k=0.6, L=20.0):
    return math.exp(-(t / L) ** k)

def offer(t, rho=0.5, tau=8.0, beta=0.03):
    return math.exp(-beta * t) * (1 - rho * math.exp(-t / tau))

if __name__ == "__main__":
    gone = 1 - surv(60)
    for t in (3, 5):
        print("   share of all swipes in the first {} s: {:.3f}".format(
            t, (1 - surv(t)) / gone))
    tau, beta, rho = 8.0, 0.03, 0.5
    ts = tau * math.log(rho * (1 + beta * tau) / (beta * tau))
    print("   offer at the best moment ({:.2f} s) vs at arrival: {:.3f}"
          .format(ts, offer(ts) / offer(0)))
    for b in (0.314, 0.6, 1.0):
        print("   hook breadth {:.3f}: casket buyers {:.3f} of viewers"
              .format(b, 1 - 20 ** (-b * b)))
#    share of all swipes in the first 3 s: 0.320
#    share of all swipes in the first 5 s: 0.413
#    offer at the best moment (7.59 s) vs at arrival: 1.284
#    hook breadth 0.314: casket buyers 0.256 of viewers
#    hook breadth 0.600: casket buyers 0.660 of viewers
#    hook breadth 1.000: casket buyers 0.950 of viewers

《文心雕龙·章句》:开篇预示中篇

原典

刘勰《文心雕龙·章句》(chinesenotes.com 所录本)

夫設情有宅,置言有位;宅情曰章,位言曰句。

啟行之辭,逆萌中篇之意;

是以搜句忌于顛倒,裁章貴于順序,斯固情趣之指歸,文筆之同致也。

(Liu, n.d.)

刘勰说,开篇的文辞要预先萌发中篇的意旨。用第一章的语言说,开头不只是吸引注意,还要让观众看到这条视频将走向哪里。程序的第一部分给出这句话在短视频中的分量:在第一章的威布尔模型下,60 秒视频的全部划走中,0.320 发生在前 3 秒,0.413 发生在前 5 秒。「裁章贵于顺序」也提醒我们:第一章的计算表明,在中段堆叠再钩并不能弥补一个没有指向的开头。

《鬼谷子·揣篇》:揣情须待其时

原典

《鬼谷子·揣篇》(Project Gutenberg 所录本)

故計國事者。則當審權量。說人主。則當審揣情。

揣情者。必以其甚喜之時。往而極其欲也。

(Guiguzi, n.d.)

《揣篇》说,游说人主要先审慎地揣度其情;揣情必须在对方最高兴的时候前往,以穷尽其欲望。撇开其政治语境,这是一条关于时机的原则:同样的请求,在对方心有旁骛时提出与在对方专注时提出,结果不同。程序的第二部分用第四章的注意力残留模型给出一个数:视频以悬念结尾时,在最佳时刻(到达后 7.59 秒)出价,比到达即出价的转化高 1.284 倍。「审揣情」也意味着测量:第六章说明,要可靠地测出付费转化的小差别,需要数万名观众。

《韩非子》:买椟还珠

原典

《韩非子·外储说左上》(Project Gutenberg 所录本)

楚人有賣其珠於鄭者,為木蘭之櫃,薰以桂椒,綴以珠玉,飾以玫瑰,輯以翡翠,鄭人買其櫝而還其珠,此可謂善賣櫝矣,未可謂善鬻珠也。

(Han, n.d.)

楚人把匣子装饰得过于华美,郑人买下匣子,退还了珍珠:这可以说善于卖匣子,却不能说善于卖珍珠。宽钩子就是一只华美的匣子。程序的第三部分以第四章的模型计算「买椟者」的比例,即被钩子吸引来、但按最窄钩子的转化率衡量不会购买的观众:现金最优宽度 0.314 时为 0.256,宽度 0.6 时为 0.660,最宽时为 0.950。最宽的钩子带来的观众,二十个里有十九个是来买匣子的。

小结

刘勰的「启行之辞,逆萌中篇之意」对应开头几秒的决定性;鬼谷子的「揣情」对应出价的时机与测量;韩非子的「买椟还珠」对应宽钩子的逆向选择。三段文字都提醒我们:吸引注意与促成交易是两件不同的事。

参考文献

Guiguzi. n.d. Guiguzi, Chapter Chuai. In Chinese; Project Gutenberg e-Text No. 7209.
Han, Fei. n.d. Hanfeizi: Waichu Shuo, Zuo Shang. In Chinese; Project Gutenberg e-Text No. 24049.
Liu, Xie. n.d. Wenxin Diaolong, Chapter 34: Zhangju. In Chinese; Chinesenotes.com Edition, Wenxin034.

Fan, Y. (2026). Short-Video Completion Hazards, Recommender Exploration Loops and Automated Acquisition Funnels: Weibull Swipe-Away Kinetics, Multi-Stage Pool Promotion Thresholds and Traffic-to-Cashflow Conversion Bounds [短视频完播率风险函数、推荐反馈环与自动化获客漏斗] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090213 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).