关于「实践与证伪:客服促成成交技能包、规则回放与预注册检验」,《私讯成交的节奏计算》给出的定量结论是什么?

结论:第八章给出七个场景的客服促成成交技能包、召回率 1.000 的停止词回放与预注册检验方案,首次回复时延试验每组需要 8,158 条询价。书中以标注框区分已发表的证据、可运行的 Python 计算(18 段程序)、原典与作者的假说。

数据口径:书中参数为示意;全部计算由书中程序生成;本书不构成个别化的法律、合规或经营建议,只讨论守规矩的成交。 第一章用动态规划切分一条 184 字的示意报价,最优为 5 个气泡,代价 46.52,一句一条的代价是 142.00。

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: 1.000

stop-word detector8 true stops, 0 missed, 2 false alarms
recall 1.000, precision 0.800
false alarm (stop anyway, ask a human): 'stop by the shop'
false alarm (stop anyway, ask a human): 'hold the shipment'
replay of 17 events from 5 customers:
blocked: after stop1
blocked: human owns chat1
blocked: over the cap1
blocked: threat wording1
deferred to 0900 3
handed to a human1
sent7
stop recorded1
查看完整验算程序
# Chapter 8. p81: the closing-skill rules run on a scripted day of chats
# Part 1 scores the stop-word detector on 16 labelled customer messages.
# Part 2 replays scripted outgoing attempts (follow-ups, payment
# reminders, replies) for five illustrative customers through the rules:
# quiet hours 21:00 to 09:00 (non-replies are deferred to 09:00), at most
# 2 follow-ups per inquiry and 2 reminders per unpaid order, stop on a
# stop word (recorded, nothing promotional afterwards), hand off to a
# human on signs of a minor, financial distress or crisis, and block any
# outgoing text with threatening wording.
import collections
import re

STOP = re.compile(r"(别|不要|不用)再?(发|联系|打扰|推)|停止发送|退订|"
                  r"^不需要[了,,。!! ]*(谢谢)?[。!!]?$|\bstop\b|"
                  r"\bunsubscribe\b", re.I)
VULN = re.compile(r"未成年|我还在上初中|家长的钱|借钱|网贷|还不上|"
                  r"活不下去|撑不住了")
THREAT = re.compile(r"起诉|上门|曝光|征信|后果自负|法律责任")

LABELLED = [("别再发了", 1, ""), ("不要再联系我", 1, ""),
            ("请停止发送", 1, ""), ("退订", 1, ""),
            ("不需要,谢谢", 1, ""), ("STOP", 1, ""),
            ("以后别打扰我", 1, ""), ("不用再推了", 1, ""),
            ("不要紧,我晚点付", 0, ""), ("我不要基础版,要高级版", 0, ""),
            ("别急,我在看", 0, ""), ("需要发票吗", 0, ""),
            ("停车场在哪", 0, ""), ("不需要发票", 0, ""),
            ("can I stop by the shop tomorrow", 0, "'stop by the shop'"),
            ("先别发货,下周再说", 0, "'hold the shipment'")]

c = collections.Counter()
for text, gold, _ in LABELLED:
    hit = 1 if STOP.search(text) else 0
    c[(gold, hit)] += 1
tp, fp, fn = c[(1, 1)], c[(0, 1)], c[(1, 0)]
print("stop-word detector: {} true stops, {} missed, {} false alarms"
      .format(tp, fn, fp))
print("  recall {:.3f}, precision {:.3f}".format(tp / (tp + fn),
                                                tp / (tp + fp)))
for text, gold, gloss in LABELLED:
    if gold == 0 and STOP.search(text):
        print("  false alarm (stop anyway, ask a human): " + gloss)

EVENTS = [  # (customer, hour, kind, text)
    ("A", 10, "in", "这个课程两档有什么区别"), ("A", 10, "reply", "两档区别如下"),
    ("A", 22, "followup", "您好,还有疑问吗"), ("A", 33, "followup", "补充一点"),
    ("A", 40, "followup", "最后确认一下"),
    ("B", 9, "in", "订单先放着"), ("B", 21, "reminder", "订单还保留着"),
    ("B", 34, "reminder", "温馨提示:订单今晚到期"),
    ("B", 35, "in", "别再发了"), ("B", 45, "reminder", "最后提醒"),
    ("C", 11, "in", "我还在上初中,可以用家长的钱买吗"),
    ("C", 11, "reply", "高级版更适合你"),
    ("D", 14, "in", "现在手头紧"),
    ("D", 15, "reminder", "再不付款将上报征信,后果自负"),
    ("E", 12, "in", "高级版能开发票吗"), ("E", 12, "reply", "可以开"),
    ("E", 23, "followup", "发票信息发我一下")]


def quiet(hour):
    h = hour % 24
    return h >= 21 or h < 9


state = collections.defaultdict(collections.Counter)
log = collections.Counter()
for who, hour, kind, text in EVENTS:
    st = state[who]
    if kind == "in":
        if STOP.search(text):
            st["stopped"] = 1
            log["stop recorded"] += 1
        if VULN.search(text):
            st["human"] = 1
            log["handed to a human"] += 1
        continue
    if st["stopped"] and kind != "reply":
        log["blocked: after stop"] += 1
        continue
    if st["human"]:
        log["blocked: human owns chat"] += 1
        continue
    if THREAT.search(text):
        log["blocked: threat wording"] += 1
        continue
    if kind in ("followup", "reminder"):
        if st[kind] >= 2:
            log["blocked: over the cap"] += 1
            continue
        st[kind] += 1
        if quiet(hour):
            log["deferred to 09:00"] += 1
    log["sent"] += 1
print("replay of {} events from 5 customers:".format(len(EVENTS)))
for key in sorted(log):
    print("  {:<26}{:>3}".format(key, log[key]))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《即时通讯气泡节奏学、已读不回破冰博弈与高客单私讯逼单母典》第 8 章「实践与证伪:客服促成成交技能包、规则回放与预注册检验」(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-145 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026145,
  author    = {Fan, Yafu},
  title     = {Instant-Messaging Bubble Rhythm, Read-No-Reply Icebreaking Games and High-Ticket Closing in Private Messages},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论