Revenue rose to $5,307.1mn in 2025, from $1,786.2mn in 2024 (p. 16). Yet the share price fell 22% after the results (p. 1).2025 年收入升至 5,307.1 百万美元,2024 年为 1,786.2 百万美元(第 16 页)。但业绩公布后,股价反而下跌 22%(第 1 页)。
The Monsters made 38% of 2025 revenue. The firm leans heavily on one IP (p. 1).The Monsters 系列贡献了 2025 年收入的 38%。公司高度依赖单一 IP(第 1 页)。
A resale market grew on Xianyu, Qiandao, StockX, eBay and Whatnot (p. 12). Some limited editions reached thousands of dollars.二手转售市场在闲鱼、千岛、StockX、eBay 和 Whatnot 上形成(第 12 页)。部分限量款售价高达数千美元。
Pop Mart chose to scale up supply instead of keeping scarcity. Resale prices then fell through 2025 (p. 12).泡泡玛特选择扩大供应,而不是维持稀缺。此后转售价格在 2025 年持续下滑(第 12 页)。
On Qiandao, a Labubu 2.0 secret figure (retail $14) fell from $124 in June 2025 to $55 in March 2026 (p. 27).在千岛上,一款 Labubu 2.0 隐藏款(零售价 14 美元)由 2025 年 6 月的 124 美元跌至 2026 年 3 月的 55 美元(第 27 页)。
Resale is a weak buying motive. Only 13% of US buyers and 16% of China buyers cite it (p. 21).转售升值并不是主要购买动机。只有 13% 的美国买家和 16% 的中国买家提到这一点(第 21 页)。
By early 2026, inventory days rose and US sales fell year on year in March (p. 12). Morgan Stanley cut its target price 11% to HK$247.到 2026 年初,存货周转天数上升,3 月美国销售同比下降(第 12 页)。摩根士丹利(Morgan Stanley)把目标价下调 11% 至 247 港元。
The case sets out four growth paths: IP portfolio, categories, entertainment and overseas (pp. 13–15). All compete for the same capital.案例列出四条增长路径:IP 组合、品类扩展、娱乐内容和海外市场(第 13–15 页)。四者争夺同一笔资本。
For our report · 对报告的意义
Ch 10: Secret figures lost more than half their value in nine months (p. 27). This supports our view that more supply lowers resale premiums.第 10 章:隐藏款在九个月内跌去一半以上的价值(第 27 页)。这支持我们「供应增加压低转售溢价」的判断。
Ch 6.1: Blind boxes and rare secret figures drive repeat buying. Only a minority buy for resale value (pp. 9, 21).第 6.1 章:盲盒和稀有隐藏款推动重复购买。只有少数人是为了转售价值而买(第 9、21 页)。
Ch 3: The 38% one-IP share is a weakness. Molly was over 60% of IP revenue in 2018, so this has happened before (pp. 1, 6).第 3 章:单一 IP 占 38% 是一项弱点。2018 年 Molly 曾占 IP 收入的 60% 以上,这种风险以前就出现过(第 1、6 页)。
Ch 10: Fakes known as "Lafufus" spread on resale markets as demand peaked (p. 11).第 10 章:需求见顶时,被称为「Lafufu」的假货在转售市场上蔓延(第 11 页)。
Limits · 局限
A company designate approved the case. Executive quotes go unchallenged.案例经公司指定人员审批。管理层的说法没有受到质疑。
USD figures are converted from CNY. They will not match CNY sources exactly.美元数字由人民币换算而来,与人民币口径的资料不会完全一致。
Resale data is one platform on three dates, with no volumes. The data stops at March 2026.转售数据只来自一个平台的三个日期,没有成交量。数据截至 2026 年 3 月。
2. World's Pop Mart: Pop Mart's IP Playbook · 世界的泡泡玛特:泡泡玛特的 IP 打法
HKU Asia Case Research Centre teaching case 香港大学亚洲案例研究中心教学案例Ch 3, 4.1, 4.4, 10
Fan, T., & Chen, H. (T.). (2026). World's Pop Mart: Pop Mart's intellectual property (IP) playbook (Ref. 26/855C) [Teaching case]. Asia Case Research Centre, The University of Hong Kong.
It shows how Pop Mart finds, grades and sells IPs, and where that system is exposed.
Pop Mart has four IP sources (p. 4). Exclusive licences give ~50% of 2025 sales and acquired IPs ~25%.泡泡玛特的 IP 有四种来源(第 4 页)。独家授权约占 2025 年销售的 50%,收购 IP 约占 25%。
In-house IPs give ~15% of sales and non-exclusive licences ~10% (p. 4).自有 IP 约占销售的 15%,非独家授权约占 10%(第 4 页)。
IPs are graded S, A, B and C (p. 5). Shelf space and marketing go to the IPs with the best market data.IP 被分为 S、A、B、C 四级(第 5 页)。货架位置和营销资源优先分给市场数据最好的 IP。
Pop Mart had 630 retail stores by Dec 2025 (p. 5). Over 2,000 roboshops act as low-cost first touchpoints (p. 6).截至 2025 年 12 月,泡泡玛特有 630 家零售店(第 5 页)。2,000 多台机器人商店(roboshop,自动售货店)是低成本的首个接触点(第 6 页)。
Pop Draw gave 40% of China online sales in 2025 (p. 14). That is ahead of Tmall at 21% and Douyin at 19%.2025 年,抽盒机(Pop Draw)贡献了中国线上销售的 40%(第 14 页),高于天猫的 21% 和抖音的 19%。
Advertising and marketing cost 3.2% of revenue in 2025 (p. 12). Customers, not paid media, spread the brand.2025 年广告和营销费用只占收入的 3.2%(第 12 页)。品牌传播主要靠顾客自发分享,而不是付费媒体。
The Monsters rose from 6% of sales in 2023 to 38% in 2025 (p. 15). Molly fell to 8%.The Monsters 占销售的比例由 2023 年的 6% 升至 2025 年的 38%(第 15 页)。Molly 则跌至 8%。
Labubu resale values fell when supply rose ~10 times to ~30 million units a month (p. 1). The share price fell ~50% from the August 2025 peak by March 2026 (p. 9).当供应增加约 10 倍、达到每月约 30 million(3,000 万)件时,Labubu 的转售价值随之下跌(第 1 页)。股价由 2025 年 8 月的高点到 2026 年 3 月下跌约 50%(第 9 页)。
For our report · 对报告的意义
Ch 4.4: Offline retail plus roboshops made 51% of 2025 sales, and online made 44% (p. 12). Pop Mart is omnichannel, not online-first.第 4.4 章:2025 年线下零售加机器人商店占销售的 51%,线上占 44%(第 12 页)。泡泡玛特走的是全渠道(omnichannel),而不是线上优先。
Ch 4.1: Pop Draw rebuilds the blind-box gamble online. It is Pop Mart's own top China online channel (pp. 6, 14).第 4.1 章:抽盒机把盲盒的博彩式体验搬到线上。它是泡泡玛特在中国最大的自有线上渠道(第 6、14 页)。
Ch 10: Premiums fell once output rose ~10 times (p. 1). One IP family also made 38% of 2025 sales (p. 15).第 10 章:产量增加约 10 倍后,溢价随即下跌(第 1 页)。同时,单一 IP 家族占 2025 年销售的 38%(第 15 页)。
Ch 3: The S/A/B/C tiers and in-house IPs such as Hirono (CNY1.7bn in 2025) suggest a repeatable system (p. 5).第 3 章:S/A/B/C 分级和 Hirono 等自有 IP(2025 年收入 CNY1.7bn,即 17 亿元人民币)说明这套体系有一定可复制性(第 5 页)。
Limits · 局限
It is a teaching case built mostly from secondary sources. The authors say they may have disguised some information (p. 1).这是一份主要基于二手资料的教学案例。作者表示可能对部分信息作了掩饰处理(第 1 页)。
The S/A/B/C tiers are described without criteria or tier members (p. 5).案例描述了 S/A/B/C 分级,但没有给出分级标准或各级包含哪些 IP(第 5 页)。
Pop Draw's 40% is of China online sales, not global online sales (pp. 12, 14). The text says offline ~55%, but the exhibit shows 51%.抽盒机的 40% 是中国线上销售的占比,不是全球线上销售的占比(第 12、14 页)。正文写线下约 55%,但附表显示为 51%。
3. Pop Mart Consumer Survey: Collectors or Trend Chasers? · 泡泡玛特消费者调查:长期收藏者还是短期潮流追随者?
Bernstein, Société Générale Group (Hu, M., & Fong, F.). (2025, December 2). Pop Mart consumer survey: Are its core consumers sustainable collectors or transient trend chasers? [Equity research report].
It is the only survey we have that splits Pop Mart's China buyers into collectors and casual buyers.
Bernstein surveyed about 850 people in China online (p. 1). No overseas buyers are covered.伯恩斯坦(Bernstein)在中国线上调查了约 850 人(第 1 页)。调查不包括海外买家。
Only 5.56% are deep collectors, and 3.19% are traders or flippers (p. 7).深度收藏者只占 5.56%,炒卖者(trader/flipper)占 3.19%(第 7 页)。
Casual or impulse buyers plus aesthetic buyers make up 68.09% (p. 7).随性或冲动型买家加上颜值型买家,合计占 68.09%(第 7 页)。
The base is new. 34.04% have never bought, and only 11.70% have bought for more than 2 years (p. 6).顾客群相当新。34.04% 从未购买过,购买超过 2 年的只有 11.70%(第 6 页)。
Spending is shallow. 68.80% spend under ¥500 a year (p. 10).消费金额偏低。68.80% 的人每年花费不到 500 元(第 10 页)。
Labubu brought in 37.94% of buyers (p. 12). Yet 13.24% plan to stop buying within 12 months (p. 16).37.94% 的买家最初是被 Labubu 吸引进来的(第 12 页)。但有 13.24% 计划在 12 个月内停止购买(第 16 页)。
75.53% have used resale platforms (p. 20). But 45.04% paid only 1–20% above retail (p. 21).75.53% 的人用过二手转售平台(第 20 页)。但 45.04% 的人最多只比零售价多付 1–20%(第 21 页)。
Blind boxes are tolerated more than loved (p. 23). 46.81% call them fun but sometimes frustrating, and 17.85% love the uncertainty.人们对盲盒更多是容忍,而不是热爱(第 23 页)。46.81% 认为「有趣但有时令人沮丧」,只有 17.85% 喜欢这种不确定性。
For our report · 对报告的意义
Ch 6: Only 5.56% are deep collectors (p. 7). The pull of the blind box is mostly design, not completing a series (p. 22).第 6 章:深度收藏者只有 5.56%(第 7 页)。盲盒的吸引力主要来自设计,而不是集齐系列(第 22 页)。
Ch 10: Resale is mainstream, but only 3.19% are flippers (pp. 7, 20). The resale premium rests on a thin layer of buyers.第 10 章:转售很普遍,但炒卖者只有 3.19%(第 7、20 页)。转售溢价建立在一小群买家身上。
Ch 4: The official app reaches only 32.74% of buyers (p. 25). Third-party e-commerce at 63.83% and stores at 63.36% reach more.第 4 章:官方 App 只覆盖 32.74% 的买家(第 25 页)。第三方电商(63.83%)和门店(63.36%)覆盖面更广。
Ch 8: Xianyu at 55.44% and Dewu at 43.26% are the main resale venues (p. 20).第 8 章:闲鱼(55.44%)和得物(43.26%)是主要的转售渠道(第 20 页)。
Limits · 局限
The broker rates the stock Underperform, so the framing leans negative.该券商给出「跑输大市」(Underperform)评级,因此解读偏向负面。
The sample is China-only and approximate. No field dates or sampling frame are given.样本只覆盖中国,人数也只是约数。报告没有给出调查日期或抽样框架。
The Dewu chart shows 43.26%, but the text says 46.23% (p. 20). We cite the chart.得物的图表显示 43.26%,正文却写 46.23%(第 20 页)。我们以图表为准。
Buechel, E. C., & Li, R. (2023). Mysterious consumption: Preference for horizontal (vs. vertical) uncertainty and the role of surprise. Journal of Consumer Research, 49(6), 987–1013. https://doi.org/10.1093/jcr/ucac039
It shows people choose mystery products to be surprised, but only when options differ by taste.
Ten US field, lab and online studies tested mystery against known options (p. 1006).作者在美国做了 10 项实地、实验室和线上研究,比较神秘选项和已知选项(第 1006 页)。
Most people chose the mystery item: 75% for ice cream and 89% for stress balls (p. 992).大多数人选了神秘商品:冰淇淋是 75%,减压球是 89%(第 992 页)。
Wanting to be surprised was the only significant predictor of choosing the mystery box (p. 994).「想要惊喜」是唯一能显著预测人们选择神秘盒的因素(第 994 页)。
A mystery snack beat a sure snack at 59%, but a classic gamble won only 28% (p. 996).神秘零食有 59% 的人选择,胜过确定的零食;传统赌博式选项只有 28% 的人选(第 996 页)。
With equal uncertainty, 73% chose an unknown flavour, 41% an unknown tea-bag count, 33% unknown money (p. 998).在不确定程度相同时,73% 的人选择未知口味,41% 选择未知茶包数量,只有 33% 选择未知金额(第 998 页)。
When outcomes looked too similar, the preference vanished: 62% chose mystery with high differentiation, 44% with low (p. 1001).如果各个结果看起来太相似,偏好就会消失:差异大时 62% 选神秘选项,差异小时只有 44%(第 1001 页)。
Pooled across studies, 70.0% prefer horizontal uncertainty, but only 41.7% prefer vertical uncertainty (p. 1001).汇总各项研究后,70.0% 的人偏好水平不确定性(horizontal uncertainty,即口味差异),但只有 41.7% 偏好垂直不确定性(vertical uncertainty,即质量或金额差异)(第 1001 页)。
For our report · 对报告的意义
Ch 6.1: A blind box sells surprise, not only the chance of a win. Regular figures differ by design (pp. 987, 994).第 6.1 章:盲盒卖的是惊喜,不只是「中奖」的机会。同一系列的常规款只在设计上不同(第 987、994 页)。
Ch 10: Hidden figures and resale prices on Xianyu and Qiandao rank designs by money. This makes the uncertainty vertical (pp. 998, 1005).第 10 章:隐藏款,以及闲鱼和千岛上的转售价格,按金钱给款式排了高低。这把不确定性变成了垂直型(第 998、1005 页)。
Ch 6.3: Each series needs designs that look clearly different. Near-identical variants kill the surprise (p. 1001).第 6.3 章:每个系列的款式要明显不同。几乎一样的款式会让惊喜消失(第 1001 页)。
Ch 6.2: Pop Draw should frame each draw as discovering a design, not as a value tier (p. 1005).第 6.2 章:Pop Draw 抽盒机应该把每次抽盒包装成发现一款设计,而不是抽中某个价值等级(第 1005 页)。
Limits · 局限
Samples are US students and online panels only. There are no Chinese buyers or collectible toys.样本只有美国学生和线上调查小组,没有中国消费者,也没有收藏类潮玩。
Choices were one-off and low-stakes, not repeat purchases.研究中的选择都是一次性、低风险的,不涉及重复购买。
There is no resale market, so prices never rank the outcomes.研究中没有二手转售市场,所以价格从来不会给结果排高低。
Mvondo, G. F. N., Jing, F., & Hussain, K. (2023). What's in the box? Investigating the benefits and risks of the blind box selling strategy. Journal of Retailing and Consumer Services, 71, Article 103189. https://doi.org/10.1016/j.jretconser.2022.103189
It links blind-box pleasures and risks to delight, and delight to brand evangelism, using Chinese buyers.
Buyers with a higher appetite for risk are more likely to buy blind boxes for delight (p. 6).越愿意冒险的消费者,越可能为了惊喜感而购买盲盒(第 6 页)。
Buyers may feel betrayed and turn against the brand when contents are worth less than the price (p. 8).如果盒内商品的价值低于售价,消费者可能觉得被背叛,并开始抵制品牌(第 8 页)。
For our report · 对报告的意义
Ch 6.1: It supports the claim that uncertainty sells through delight. It credits Pop Mart with popularising blind boxes in China (p. 1).第 6.1 章:这篇论文支持「不确定性通过惊喜感促成购买」的观点。它认为是泡泡玛特让盲盒在中国流行起来的(第 1 页)。
Ch 6.2: Adventure and social benefits match unboxing, chasing hidden figures and the Pop Draw app.第 6.2 章:冒险利益和社交利益,对应拆盒、追隐藏款和 Pop Draw 抽盒机的玩法。
Ch 6.4: Risk-seeking buyers are drawn in more, which supports our gambling-risk section (p. 6).第 6.4 章:爱冒险的消费者更容易被吸引,这支持我们关于赌博风险的分析(第 6 页)。
Ch 8: It gives us coding labels: delight, repeat buying, recommending, and anti-brand reactions to poor value (pp. 6, 8).第 8 章:它为我们的顾客之声(Voice of the Customer)编码提供了标签:惊喜、复购、推荐,以及因性价比差而抵制品牌(第 6、8 页)。
Limits · 局限
The survey excluded collectible toy figures, so it does not measure Pop Mart's core product (p. 4).调查排除了收藏类小型潮玩,所以没有测量泡泡玛特的核心产品(第 4 页)。
The sample is purposive or convenience-based, not random (pp. 4, 8).样本是目的抽样或便利抽样,不是随机抽样(第 4、8 页)。
It is one cross-sectional survey and does not compare age or gender groups (p. 8).这只是一次横断面调查,也没有比较不同年龄或性别群体(第 8 页)。
6. UI dark patterns and where to find them · 界面暗黑模式在哪里
Peer-reviewed conference paper 同行评审会议论文Ch 6.2, 6.3, 7
Di Geronimo, L., Braz, L., Fregnan, E., Palomba, F., & Bacchelli, A. (2020). UI dark patterns and where to find them: A study on mobile applications and user perception. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (CHI '20) (pp. 1–14). ACM. https://doi.org/10.1145/3313831.3376600
It shows dark patterns are in almost every popular app, and most users miss them.
Study 1 covered 240 trending free Android apps, 30 from each of 8 categories (p. 3).研究 1 覆盖了 240 个热门免费安卓应用,来自 8 个类别,每类 30 个(第 3 页)。
Researchers used each app for ten minutes, producing 40 hours of coded video (p. 4).研究者每个应用使用十分钟,共得到 40 小时的视频供编码(第 4 页)。
95% of the apps contained at least one dark pattern (p. 5).95% 的应用至少含有一种暗黑模式(dark pattern)(第 5 页)。
The team found 1,787 dark patterns, an average of 7.4 per app (p. 5).研究共发现 1,787 个暗黑模式,平均每个应用 7.4 个(第 5 页)。
False Hierarchy appeared in 61% of apps, and Preselection in 60% (p. 5).虚假层级(False Hierarchy,有利于应用的选项更大更显眼)出现在 61% 的应用中,预选(Preselection)出现在 60% 的应用中(第 5 页)。
Study 2 surveyed 589 people from 46 countries using app videos (pp. 7–8).研究 2 用应用视频,在线调查了来自 46 个国家的 589 人(第 7–8 页)。
In apps with dark patterns, 55% of answers missed them and 20% were unsure (p. 8).面对含暗黑模式的应用,55% 的回答没有发现它们,20% 不确定(第 8 页)。
Only 14% spotted the insurance that the shopping app Romwe pre-added at checkout (p. 8).购物应用 Romwe 在结账时默认加入保险,只有 14% 的人发现了(第 8 页)。
For our report · 对报告的意义
Ch 7: Our audit cannot rely on user complaints, because most users do not notice dark patterns (p. 8).第 7 章:我们的审查不能只看用户投诉,因为大多数用户察觉不到暗黑模式(第 8 页)。
Ch 7: We reuse their task script, from sign-up to checkout, and check for pre-ticked add-ons (p. 4).第 7 章:我们沿用论文的任务脚本,从注册走到结账,并检查是否有默认勾选的附加项(第 4 页)。
Ch 6.3: Countdowns and limited-offer pressure count as dark patterns. Whether Pop Mart's timers are artificial is ours to test (p. 3).第 6.3 章:倒计时和「限时优惠」施压都属于暗黑模式。泡泡玛特的倒计时是否人为制造紧迫感,需要我们自己检验(第 3 页)。
Ch 6.2: Intermediate Currency and Gamification are labels for Pop Draw points and repeat-draw loops (p. 3).第 6.2 章:中间货币(Intermediate Currency)和游戏化(Gamification)这两个标签,可用来描述 Pop Draw 抽盒机的积分和反复抽盒循环(第 3 页)。
Limits · 局限
The sample is free US Android apps, with no Chinese apps or mini-programs (p. 3).样本只有美国的免费安卓应用,没有中国应用,也没有小程序(第 3 页)。
Only the first ten minutes were coded, with no purchases, so post-purchase patterns were missed (p. 4).只编码了前十分钟,也没有实际付款,所以漏掉了购买后的暗黑模式(第 4 页)。
Users watched 30-second videos instead of using the apps, which may lower detection (p. 10).受访者看的是 30 秒视频,而不是亲自使用应用,这可能让发现率偏低(第 10 页)。
7. Live streaming commerce and purchase intention · 直播电商与购买意愿
Lu, B., & Chen, Z. (2021). Live streaming commerce and consumers' purchase intention: An uncertainty reduction perspective. Information & Management, 58, Article 103509. https://doi.org/10.1016/j.im.2021.103509
It shows live-stream hosts sell mainly by building trust, which lowers shoppers' doubts.
The survey had 535 valid Chinese live-stream shoppers, collected 18 June to 5 July 2020 (p. 8).调查共有 535 份有效样本,都是中国的直播购物者,收集时间为 2020 年 6 月 18 日至 7 月 5 日(第 8 页)。
Feeling physically similar to the host lowers doubts about fit (β = −0.153) (p. 9).觉得自己和主播身材相似,会减少对合身程度的疑虑(β = −0.153)(第 9 页)。
Feeling that the host shares one's values raises trust in the host (β = 0.559) (p. 9).觉得主播和自己价值观相近,会提升对主播的信任(β = 0.559)(第 9 页)。
Trust lowers uncertainty about both fit and quality (p. 9).信任能同时减少对合身程度和质量的不确定感(第 9 页)。
Both kinds of uncertainty lower purchase intention, while trust raises it (β = 0.289) (p. 9).两种不确定感都会降低购买意愿,而信任会提升购买意愿(β = 0.289)(第 9 页)。
Trust has the largest total effect on purchase intention (0.514) (p. 9).在所有因素中,信任对购买意愿的总效应最大(0.514)(第 9 页)。
Value similarity builds trust more strongly on Douyin-type platforms (p. 12).在抖音这类平台上,价值观相似对建立信任的作用更强(第 12 页)。
For our report · 对报告的意义
Ch 4: Douyin and TikTok live streams sell through trust in the host, not only through price (p. 12).第 4 章:抖音和 TikTok 的直播靠观众对主播的信任来卖货,不只是靠价格(第 12 页)。
Ch 4: A host can show figure quality, but not which figure is in a sealed box. This is our inference.第 4 章:主播能展示公仔的质量,却无法展示密封盒里是哪一款。这是我们的推论。
Ch 6.1: The paper treats uncertainty as a barrier. Pop Mart keeps design uncertainty on purpose and reduces only quality doubts.第 6.1 章:这篇论文把不确定性看作购买障碍。泡泡玛特则有意保留款式的不确定性,只减少对质量的疑虑。
Ch 10: A trusted seller lowers quality doubts, which matters on Xianyu and Qiandao where fakes circulate (p. 9).第 10 章:可信的卖家能减少对质量的疑虑。这在假货流通的闲鱼和千岛上尤其重要(第 9 页)。
Limits · 局限
It covers clothes and cosmetics only (p. 13).研究只涉及服装和化妆品(第 13 页)。
The sample is Chinese only, which limits use in other markets (p. 13).样本只有中国消费者,难以直接推广到其他市场(第 13 页)。
It is a self-reported 2020 survey that measures intention, not actual buying (p. 8).这是 2020 年的自填式问卷,测量的是购买意愿,不是实际购买(第 8 页)。
8. Parallel platformization of Douyin and TikTok · 抖音与 TikTok 的平行平台化
Kaye, D. B. V., Chen, X., & Zeng, J. (2021). The co-evolution of two Chinese mobile short video apps: Parallel platformization of Douyin and TikTok. Mobile Media & Communication, 9(2), 229–253. https://doi.org/10.1177/2050157920952120
It shows Douyin and TikTok share one interface but differ in selling features and rules.
Two researchers walked through both apps side by side in August and September 2019 (p. 6).两名研究者在 2019 年 8 月和 9 月,并排使用两个应用做了系统走查(walkthrough)(第 6 页)。
The interfaces are almost identical, with the same logo and a central create button (p. 7).两个应用的界面几乎一样,标志相同,中间都有一个创作按钮(第 7 页)。
Local differences sit in details, such as Douyin's skin-lightening default filter (p. 7).本地差异体现在细节上,例如抖音默认的相机滤镜会让肤色变白(第 7 页)。
Both apps open on an endless, algorithm-ranked "For You" feed that plays automatically (p. 8).两个应用一打开都是「为你推荐」信息流:由算法排序、自动播放、永远刷不到底(第 8 页)。
Douyin has no web browser version, which keeps users inside the app (p. 10).抖音没有网页版,用户只能留在应用里(第 10 页)。
Live streaming needs 1,000 followers on TikTok, but 50,000 plus real-name checks on Douyin (p. 11).在 TikTok 开直播需要 1,000 名粉丝;在抖音则需要 50,000 名粉丝,还要实名认证(第 11 页)。
Douyin's daihuo feature links product icons inside live streams; TikTok had none in 2019 (pp. 13–14).抖音有「带货」(daihuo)功能,在直播中嵌入可点击的商品图标;2019 年的 TikTok 没有这项功能(第 13–14 页)。
For our report · 对报告的意义
Ch 4: Pop Mart's Douyin store works as an in-app shop, with clickable products inside live streams (pp. 10, 13).第 4 章:泡泡玛特的抖音店铺是应用内商店,直播中可直接点击商品(第 10、13 页)。
Ch 4: TikTok Shop may not copy the Douyin playbook; in 2019 the model was less effective abroad (p. 15).第 4 章:TikTok Shop 未必能照搬抖音的打法。2019 年这套模式在海外效果较差(第 15 页)。
Ch 5: The endless "For You" feed is where blind-box unboxing clips get discovered (pp. 8, 18).第 5 章:无限滚动的「为你推荐」信息流,是盲盒拆盒视频被发现的地方(第 8、18 页)。
Ch 9: Our benchmark can compare Pop Mart's Douyin and TikTok pages by feature, money-making and rules (p. 16).第 9 章:我们的对标分析可以从功能、变现方式和平台规则三方面,比较泡泡玛特的抖音和 TikTok 页面(第 16 页)。
Limits · 局限
Data are from 2019, before TikTok Shop and today's Douyin mall features.数据来自 2019 年,早于 TikTok Shop 和现在的抖音商城功能。
It is a qualitative walkthrough by two researchers, with no user data or sales figures.这是两名研究者做的定性走查,没有用户数据,也没有销售数字。
Page numbers come from the author manuscript, not the journal's pages 229–253.页码来自作者手稿版,不是期刊正式版的第 229–253 页。
9. AI–human hybrids for marketing research · 人机协作的营销研究
Peer-reviewed journal article 同行评审期刊论文Ch 2, 8, 13
Arora, N., Chakraborty, I., & Nishimura, Y. (2025). AI–human hybrids for marketing research: Leveraging large language models (LLMs) as collaborators. Journal of Marketing, 89(2), 43–70. https://doi.org/10.1177/00222429241276529
It backs our method of letting an LLM code reviews while a human checks the work.
The authors replicated two 2019 studies from a Fortune 500 food company using GPT-4 (p. 44).作者用 GPT-4 重做了一家《财富》500 强食品公司在 2019 年做的两项研究(第 44 页)。
The LLM and humans highlighted similar key sentences, with average cosine similarity of .78 (p. 55).LLM 和人工标出的关键句很相似,平均余弦相似度(cosine similarity)为 .78(第 55 页)。
Humans highlighted 35 sentences on average, but the LLM only 19 (p. 55).人工平均标出 35 句,LLM 只标出 19 句(第 55 页)。
The LLM drops sentences that repeat an idea it already has (p. 55).LLM 会略去重复同一观点的句子(第 55 页)。
Coding human answers, the LLM recovered 96% of human themes and found 23% new ones (p. 55).给人工收集的回答编码时,LLM 找回了 96% 的人工主题,还发现了 23% 的新主题(第 55 页)。
None of the five expert judges chose a human-only or LLM-only summary as best (p. 55).五位专家评审中,没有一位认为纯人工或纯 LLM 的总结是最好的(第 55 页)。
In a 605-person survey replication, the LLM got answer direction right but missed the spread (pp. 58, 66).在重做 605 人问卷时,LLM 能判断回答的方向,但抓不准回答的分布(第 58、66 页)。
LLMs carry gender, race and cultural biases, so a human must stay in the loop (p. 67).LLM 带有性别、种族和文化偏见,所以必须有人参与把关(human in the loop)(第 67 页)。
For our report · 对报告的意义
Ch 8: Our LLM coding of 5,938 reviews matches the "human generated–LLM analyzed" condition, which kept 96% of themes (p. 55).第 8 章:我们用 LLM 给 5,938 条评论编码,对应论文里「人工收集、LLM 分析」的条件,那个条件保留了 96% 的主题(第 55 页)。
Ch 8: We hand-code a random sample and report our own agreement score. The paper gives none.第 8 章:我们人工编码一个随机样本,并报告自己的一致性指标。论文本身没有提供这类指标。
Ch 8: Theme counts must come from per-review labels, because LLM summaries drop repeats (p. 55).第 8 章:主题出现次数要用逐条评论的标签来算,因为 LLM 总结会略去重复内容(第 55 页)。
Ch 8: Tmall, Xiaohongshu and Xianyu reviews are in Chinese, and culture-specific work needs human judgement (p. 44).第 8 章:天猫、小红书和闲鱼的评论都是中文,涉及特定文化的研究需要人工判断(第 44 页)。
Limits · 局限
Both studies come from one US food company, in English, using GPT-4 (p. 44).两项研究都来自同一家美国食品公司,用英文,用 GPT-4(第 44 页)。
The expert check used five questions and five judges, so results are largely qualitative (p. 54).专家评估只有五个问题、五位评审,所以结果基本是定性的(第 54 页)。
It tests interview answers, not short online reviews.论文测试的是访谈回答,不是简短的网络评论。
10. Resale's next chapter · 二手转售的新篇章
Consulting report 咨询公司报告Ch 6.1, 10, 11
Moizant, F., Krueger, F., Martinez-Pardo, C., Dupreelle, P., Vitrani, J., Gardet, C., Wone, D., Hebrard-Lemaire, A., Baz, D., Dietze, E., & Lasa Aresti, M. (2025, October 9). Resale's next chapter: How fashion and luxury brands can win in the secondhand market [Consulting report]. Boston Consulting Group & Vestiaire Collective.
It sizes the resale market and shows resale value and authentication shape new-item buying.
The secondhand market is $210–220 billion today (p. 3).二手市场目前规模为 2,100–2,200 亿美元($210–220 billion)(第 3 页)。
It should reach $320–360 billion by 2030 (p. 3).预计到 2030 年将达到 3,200–3,600 亿美元($320–360 billion)(第 3 页)。
It grows about 10 percent a year, three times faster than firsthand (p. 3).二手市场每年增长约 10%,速度是新品市场的三倍(第 3 页)。
Nearly 80 percent cite affordability as a key reason to buy secondhand (p. 5).近 80% 的人把价格实惠列为购买二手的主要原因(第 5 页)。
For 66 percent, resale was how they first found or bought a brand, up from 59 percent in 2022 (p. 6).66% 的人是通过二手市场第一次发现或购买某个品牌的,2022 年这一比例是 59%(第 6 页)。
Gen Z values "the thrill of the hunt" and sold-out items more, by 7 points (p. 6).Z 世代更看重「搜寻的刺激」和已售罄的单品,比平均高 7 个百分点(第 6 页)。
25 percent of Americans consider resale value when buying new, versus 19 percent in Europe (p. 8).25% 的美国人在买新品时会考虑转售价值,欧洲是 19%(第 8 页)。
Authentication is the most valued Digital Product Passport feature, at 70 percent when buying (p. 12).正品鉴定是数字产品护照(Digital Product Passport,DPP)最受重视的功能,购买时有 70% 的人看重(第 12 页)。
For our report · 对报告的意义
Ch 10: If 25 percent of US buyers weigh resale value, resale value can support primary prices (p. 8).第 10 章:如果 25% 的美国买家会考虑转售价值,那么转售价值能支撑新品价格(第 8 页)。
Ch 6.1: "The thrill of the hunt" for limited items matches chasing hidden figures. This is our analogy (p. 6).第 6.1 章:追逐限量单品的「搜寻的刺激」,和追隐藏款很相似。这是我们的类比(第 6 页)。
Ch 10: Authentication tops DPP features, which backs a digital ID or anti-fake tag for figures (p. 12).第 10 章:正品鉴定是 DPP 最受重视的功能,这支持给公仔加数字身份或防伪标签(第 12 页)。
Ch 11: Rolex Certified Pre-Owned is a model for official Pop Mart resale (p. 10).第 11 章:劳力士官方认证二手(Rolex Certified Pre-Owned)可作为泡泡玛特官方二手转售的参考(第 10 页)。
Limits · 局限
A resale platform co-wrote the report, so its stance favours resale.报告由一家二手平台合著,立场偏向二手转售。
The 7,800 respondents are resale-engaged, fashion-conscious and affluent (p. 16).7,800 名受访者都参与二手交易、关注时尚,而且比较富裕(第 16 页)。
It focuses on the US and Europe, with no China or toy data.报告聚焦美国和欧洲,没有中国或潮玩的数据。
11. How Pop Mart Won Young Customers · 泡泡玛特如何赢得年轻顾客
Harvard Business Review practitioner article 《哈佛商业评论》实务文章Ch 4.4, 5, 8, 10
Li, Y. (2025, July 10). How Pop Mart won young customers in a fragmented attention economy. Harvard Business Review (Reprint H08T6S).
It explains how Pop Mart uses fast feedback, identity and fan language to win digital-native buyers.
Short videos split young people's attention (p. 1). The author says big-budget product development is becoming less efficient.短视频让年轻人的注意力日益碎片化(第 1 页)。作者认为,大预算的产品开发正变得不再高效。
Pop Mart tests designs, reads live feedback and moves resources to winners (p. 3). The author compares this to TikTok's algorithm.泡泡玛特先测试设计,再读取实时反馈,并把资源转向表现好的产品(第 3 页)。作者把这比作 TikTok 的算法。
Lisa and Rihanna posted Labubu without being paid (p. 4). Pop Mart then created a "soft vinyl plush" category.Lisa 和 Rihanna 在没有收费的情况下发布了 Labubu(第 4 页)。泡泡玛特随后推出了「搪胶毛绒」(soft vinyl plush)品类。
Overseas, Shopee and TikTok feedback data guided localisation (p. 4). Supply chain work allowed a 30-fold production rise within a year (p. 5).在海外,Shopee 和 TikTok 的反馈数据指导了本地化(第 4 页)。供应链改进让产量在一年内提高了 30 倍(第 5 页)。
Limited and hidden editions meet emotional and identity needs (p. 6). Unboxing posts then bring in new customers.限量款和隐藏款满足了情感和身份认同需求(第 6 页)。开箱帖又进一步带来新顾客。
Japanese rivals use distributors, but Pop Mart runs its own flagship stores (pp. 6–7). Instagram and Reddit communities amplify them.日本竞争对手依赖分销商,泡泡玛特则自营旗舰店(第 6–7 页)。Instagram 和 Reddit 社群进一步放大了门店的影响。
Fan terms such as 端盒, 拆盒 and 娃友 build a sense of belonging (p. 8).「端盒」(整箱买以抽中目标款)、「拆盒」和「娃友」等粉丝用语营造了归属感(第 8 页)。
The article puts 2024 revenue at $1.8 billion, with nearly 40% from outside mainland China (p. 2).文章称 2024 年收入为 $1.8 billion(18 亿美元),其中近 40% 来自中国内地以外(第 2 页)。
For our report · 对报告的意义
Ch 5: Celebrity spread was organic, and Pop Mart amplified it without paying (p. 4).第 5 章:名人带动的传播是自发的,泡泡玛特没有付费,只是顺势放大(第 4 页)。
Ch 4.4: Stores create content, and online communities feed store visits (p. 7). This supports our online–offline loop.第 4.4 章:门店产生内容,线上社群又带动到店(第 7 页)。这支持我们的线上线下闭环论点。
Ch 8: The fan words 端盒, 拆盒 and 娃友 give us search terms for Voice of the Customer coding (p. 8).第 8 章:「端盒」「拆盒」「娃友」等粉丝用语,可作为顾客之声(Voice of the Customer)编码的检索词(第 8 页)。
Ch 10: The article praises scarcity but never mentions the resale market it creates.第 10 章:文章称赞稀缺策略,却完全没有提到稀缺带来的转售市场。
Limits · 局限
It names no data or method. Its claims cannot be checked.文章没有说明数据或方法,其论点无法核实。
It was written in July 2025, near the peak. It cannot see the 2026 slowdown.文章写于 2025 年 7 月,接近热潮顶点,看不到 2026 年的放缓。
It is positive by design and covers no risks. The revenue and overseas figures have no named source.文章立场正面,没有讨论任何风险。收入和海外占比数字也没有注明来源。
12. Pop Mart FY2025 Annual Results Announcement · 泡泡玛特 2025 年度业绩公告
Pop Mart International Group Limited. (2026, March 25). Annual results announcement for the year ended 31 December 2025 [Announcement]. HKEXnews. https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0325/2026032500285.pdf
The company's own audited figures. We use them to check every number from the cases and the brokers.
Revenue was RMB37,120.1 million in 2025, up 184.7% year on year (p. 21).2025 年收入为人民币 37,120.1 百万元,同比增长 184.7%(第 21 页)。
Gross profit margin rose from 66.8% to 72.1%. The company credits more overseas sales (p. 35).毛利率由 66.8% 升至 72.1%。公司归因于海外销售占比上升(第 35 页)。
The PRC fell to 56.2% of revenue, from 68.2% a year earlier (p. 27).中国内地占收入的比例由 68.2% 降至 56.2%(第 27 页)。
Pop Draw (抽盒机) earned RMB3,423.5 million, up 207.4%. That is 16.4% of PRC revenue (pp. 28–29).抽盒机(Pop Draw)收入为人民币 3,423.5 百万元,增长 207.4%,占中国内地收入的 16.4%(第 28–29 页)。
The company calls Pop Draw a "gamified channel experience" and has added it to its global app (p. 25).公司把抽盒机称为「游戏化的渠道体验」(gamified channel experience),并已加入海外 App(第 25 页)。
THE MONSTERS (home of LABUBU) made 38.1% of revenue, up from 23.3% (p. 33).THE MONSTERS 系列(LABUBU 所在系列)占收入 38.1%,上年为 23.3%(第 33 页)。
Mainland members reached 72.58 million. Members made 93.7% of sales, with a 55.7% repeat purchase rate (p. 25).内地注册会员达 7,258 万(72.58 million)。会员贡献 93.7% 的销售,复购率为 55.7%(第 25 页)。
For our report · 对报告的意义
Ch 3 and 4: the channel split (Pop Draw, Tmall, Douyin, TikTok) is the base for our channel map.第 3、4 章:抽盒机、天猫、抖音、TikTok 的渠道拆分,是我们渠道架构图的基础。
Ch 6 and 10: one IP at 38.1% of revenue means a LABUBU resale fall hits the whole group.第 6、10 章:单一 IP 占收入 38.1%,LABUBU 二手价格下跌会影响整个集团。
Limits · 局限
Only the accounts are audited. The business review is the company's own positive view.只有财务报表经过审计。业务回顾是公司自己的正面说法。
Nothing on resale prices, scalpers or fakes. Member data cover the Chinese mainland only (p. 25).没有二手价格、黄牛或假货的内容。会员数据只覆盖中国内地(第 25 页)。
13. Pop Mart: 20+ Leading Indicators · 泡泡玛特:20 多项领先指标
Sell-side equity research report (high-frequency data tracker) 卖方股票研究报告(高频数据追踪)Ch 4, 6, 10
Bernstein (Hu, M., & Zhang, K.). (2026, April 21). Pop Mart: YTD high frequency data – 20+ leading indicators point to no fundamental inflection [Equity research report].
It is our strongest evidence that ordinary Labubu resale prices fell below retail by March 2026.
Bernstein tracks 20+ third-party sources to March 2026 (p. 1). It says no indicator shows a fundamental turn.伯恩斯坦(Bernstein)追踪 20 多个第三方数据源,截至 2026 年 3 月(第 1 页)。它认为没有任何指标显示基本面出现拐点。
US offline card transactions fell 18.7% YoY month-to-date in March 2026 (p. 1). They were +90% in January.2026 年 3 月截至当时,美国线下刷卡交易同比下降 18.7%(第 1 页)。1 月时还是增长 90%。
86% of tracked Labubu SKUs (stock-keeping units) now trade below retail price (p. 1).在追踪的 Labubu 款式(SKU)中,86% 的转售价已低于零售价(第 1 页)。
In March, China resale prices were down 35% YoY and volumes down 51% YoY (p. 6).3 月,中国转售价格同比下跌 35%,成交量同比下跌 51%(第 6 页)。
Secret SKUs still carry 123–409% premiums (p. 11). Non-secret SKUs trade up to 30% below retail.隐藏款(secret SKU)仍有 123–409% 的溢价(第 11 页)。非隐藏款则最多比零售价低 30%。
For the first time, a major series' secret SKUs fell (p. 15). Twinkle Twinkle Crush on You dropped 23% MoM.主要系列的隐藏款首次出现下跌(第 15 页)。Twinkle Twinkle Crush on You 环比下跌 23%。
Falling prices did not lift resale volume (p. 17). Bernstein reads this as buyers and sellers both leaving.价格下跌并没有带动转售成交量(第 17 页)。伯恩斯坦认为这说明买家和卖家都在离场。
China online platform GMV still grew 117% YoY in March 2026 (p. 6). The signal is mixed by channel.2026 年 3 月,中国线上平台商品交易总额(GMV)仍同比增长 117%(第 6 页)。不同渠道的信号并不一致。
For our report · 对报告的意义
Ch 10: Most ordinary Labubu SKUs sell below retail (p. 1). Pop Mart has not held the resale price.第 10 章:大多数普通款 Labubu 的售价已低于零售价(第 1 页)。泡泡玛特没有守住转售价格。
Ch 10: The resale market is two-tier (p. 11). Only rare secret SKUs keep a premium, and even these began to slip (p. 15).第 10 章:转售市场已分成两层(第 11 页)。只有稀有隐藏款保有溢价,而且连它们也开始下滑(第 15 页)。
Ch 6: Hidden-edition odds hold value, but common pulls lost theirs. Gamification alone does not protect price.第 6 章:隐藏款的稀有概率仍能保值,普通款却失去了价值。单靠游戏化(gamification)无法守住价格。
Ch 4: Google Trends fell 91% from the June 2025 peak (p. 6). App DAU fell 43% from its peak (p. 25).第 4 章:Google Trends 搜索热度较 2025 年 6 月高峰下跌 91%(第 6 页)。App 日活跃用户(DAU)较高峰下跌 43%(第 25 页)。
Limits · 局限
It is a bearish sell-side report rated Underperform. The tracker is built to test a negative thesis.这是一份看淡的卖方报告,评级为「跑输大市」(Underperform)。整套追踪是为检验负面论点而设计的。
The data are third-party proxies, not company sales. Resale samples cover selected SKUs only (p. 6).数据是第三方替代指标,不是公司销售数据。转售样本只涵盖部分款式(第 6 页)。
Bernstein changed its comparison method mid-series (p. 6). The data end in March 2026.伯恩斯坦在数据序列中途更改了对比方法(第 6 页)。数据截至 2026 年 3 月。
14. Selling secondhand products through an online platform with blockchain · 借助区块链通过在线平台销售二手商品
Analytical journal article (game-theory pricing model with numerical examples; no empirical data). Read from the Europe PMC full text, so locations are section headings, not pages. 分析型期刊论文(博弈论定价模型 + 数值算例;没有实证数据)。本文据 Europe PMC 全文阅读,位置以章节标题代替页码。Ch 6.3, 9, 10, 11
Shen, B., Xu, X., & Yuan, Q. (2020). Selling secondhand products through an online platform with blockchain. Transportation Research Part E: Logistics and Transportation Review, 142, Article 102066. https://doi.org/10.1016/j.tre.2020.102066
Owners consign used goods to one online platform, which competes with a brand selling new goods.
Without blockchain, buyers and the brand only guess the used item's quality; with it, everyone sees the true quality.没有区块链时,买家和品牌只能猜二手商品的质量;有了区块链,所有人都能看到真实质量。
Whether blockchain pays depends on product uniqueness and true quality: it suits unique, high-quality goods such as luxury items.区块链是否划算,取决于产品独特性和真实质量:它适合独特且高质量的商品,例如奢侈品。
With blockchain, a brand that runs its own resale platform (horizontal integration) always earns more total profit.使用区块链时,品牌自营二手平台(横向整合)总能获得更高的总利润。
The model assumes a used item never costs more than a new one, so it leaves out limited editions.模型假设二手价不高于新品价,因此排除了限量版商品。
The paper studies the quality-disclosure effect of blockchain for secondhand goods (Abstract).论文研究区块链对二手商品的质量披露效应(摘要)。
Contributors consign used goods to an online platform, which resells them (Abstract).出售者(contributor)把二手商品寄售(consign)给在线平台,由平台转卖(摘要)。
Pricing strategy changes with the revenue share in the consignment contract (Abstract).定价策略随寄售合同中的收入分成比例而变化(摘要)。
With blockchain, horizontal integration raises total supply-chain profit more (Abstract).使用区块链时,横向整合(horizontal integration)更能提升供应链总利润(摘要)。
A "win-win-win" outcome is possible when low-uniqueness goods are sold (Abstract).销售低独特性商品时,可以实现「三赢」结果(摘要)。
For our report · 对报告的意义
Ch 9: The paper explains why fakes thrive on Xianyu 闲鱼: buyers cannot judge quality from photos (Background and motivation).第 9 章:论文解释了假货为何在闲鱼泛滥:买家无法凭照片判断质量(背景与动机)。
Ch 6.3: Hidden figures 隐藏款 and Labubu look like high-uniqueness goods, where blockchain helps both brand and platform (Equilibrium analysis). This mapping is ours.第 6.3 章:隐藏款和 Labubu 类似高独特性商品,在这种情形下区块链对品牌和平台都有利(均衡分析)。这一对应是我们的判断。
Ch 10: Proven quality can push resale prices up, so verification may raise, not cool, Labubu resale prices (Welfare analysis).第 10 章:证明质量可能推高二手价,因此验真可能抬高而非压低 Labubu 二手价(福利分析)。
Ch 11: With blockchain, a brand-run resale channel always earns more total profit than a separate platform (With the use of blockchain). This argues for an official Pop Mart 泡泡玛特 channel over leaving resale to Qiandao 千岛.第 11 章:有区块链时,品牌自营二手渠道的总利润总高于独立平台(有区块链)。这支持泡泡玛特建立官方渠道,而非把二手交易交给千岛。
Limits · 局限
It is a pure theory model with no sales data, so it shows directions, not sizes.这是纯理论模型,没有销售数据,只能说明方向,不能说明大小。
It caps used prices at the new price, which rules out Pop Mart's main resale case: hidden figures above retail.它把二手价上限设为新品价,排除了泡泡玛特最主要的二手情形:隐藏款溢价高于零售价。
Quality here means wear and condition; for sealed blind boxes 盲盒, authenticity matters more than wear.这里的质量指磨损和成色;对未拆封的盲盒而言,真伪比成色更重要。
15. A Survey on Aspect-Based Sentiment Analysis: Tasks, Methods, and Challenges · 基于方面的情感分析综述:任务、方法与挑战
Literature survey in natural language processing (NLP). We read the arXiv v2 version (22 PDF pages, dated 6 Nov 2022); page numbers below are PDF pages 1–22, not the journal pages 11019–11038. 自然语言处理(natural language processing, NLP)领域的文献综述。我们读的是 arXiv v2 版本(共 22 页 PDF,日期 6 Nov 2022);下文页码均为 PDF 页码 1–22,而非期刊页码 11019–11038。Ch 2, 8, 13
Zhang, W., Li, X., Deng, Y., Bing, L., & Lam, W. (2023). A survey on aspect-based sentiment analysis: Tasks, methods, and challenges. IEEE Transactions on Knowledge and Data Engineering, 35(11), 11019–11038. https://doi.org/10.1109/TKDE.2022.3230975
Aspect-based sentiment analysis (ABSA) finds what people like or dislike about each part of a product, not just the whole review.
The survey builds everything on four sentiment elements: aspect term, aspect category, opinion term and sentiment polarity.整篇综述建立在四个情感要素之上:方面词(aspect term)、方面类别(aspect category)、观点词(opinion term)和情感极性(sentiment polarity)。
It sorts tasks into single tasks (one element) and compound tasks (pairs, triplets or quads of elements).它把任务分为单一任务(只预测一个要素)和复合任务(预测要素的二元组、三元组或四元组)。
Pre-trained language models such as BERT lifted results, but the hardest task still scores only about 40% F1.BERT 等预训练语言模型大幅提高了效果,但最难的任务 F1 仍只有约 40%。
Classic sentiment analysis gives one label to a whole sentence or document (p. 1).传统情感分析只给整句或整篇文档打一个情感标签(p. 1)。
That assumes one sentiment about one topic, which real reviews often break (p. 1).这假设文本只对一个话题表达一种情感,而真实评论常常不是这样(p. 1)。
ABSA moves the target to an entity or one aspect of it, such as a product's price or size (p. 1).ABSA 把分析对象换成实体或实体的某个方面,例如商品的价格或尺寸(p. 1)。
The authors name e-commerce reviews as a main use: they help improve products and marketing (p. 1).作者把电商评论列为主要应用:它能帮助改进产品和营销活动(p. 1)。
Early work predicted one element at a time; recent work predicts several linked elements together (p. 1).早期研究一次只预测一个要素;近年的研究同时预测多个相互关联的要素(p. 1)。
For our report · 对报告的意义
Ch 8: In one review, "the shake button" is the aspect term, the draw screen is the aspect category, "rigged" is the opinion term, and polarity is negative (pp. 2, 7).第 8 章:在一条评论中,“摇一摇按钮”是方面词,抽盒界面是方面类别,“被操控(rigged)”是观点词,情感极性为负面(pp. 2, 7)。
Ch 8: Our other categories (payment, shipping, hidden-edition odds, resale price) work the same way; "it's a scam" with no target is an implicit aspect (p. 2).第 8 章:我们的其他类别(支付、物流、隐藏款概率、二手转售价)同理;没有对象的“it's a scam”属于隐含方面(p. 2)。
Ch 2: Coding 5,938 reviews into a fixed category list with one polarity each is ACSA; if we also keep the quoted phrase, it moves towards ASQP (pp. 9, 12).第 2 章:把 5,938 条评论归入固定类别清单、并给每个类别一个极性,这就是 ACSA;如果同时保留引用短语,就接近 ASQP(pp. 9, 12)。
Ch 2: The survey supports our choice: ACSA handles implied aspects and is widely used in industry (p. 9).第 2 章:综述支持我们的选择:ACSA 能处理隐含方面,并在业界广泛使用(p. 9)。
Limits · 局限
The v2 text is dated 6 Nov 2022 and never discusses prompting chat LLMs such as GPT-4 to code reviews.v2 版本日期为 6 Nov 2022,完全没有讨论用 GPT-4 等对话式大语言模型按提示给评论编码。
It is written for NLP researchers: it compares model architectures and benchmark F1, not business insight.它是写给 NLP 研究者看的:比较的是模型结构和基准 F1,而不是商业洞察。
Benchmarks are short English laptop and restaurant sentences, unlike long, slang-heavy, mixed-language Pop Mart reviews (p. 5).基准数据多为简短的英文笔记本电脑和餐厅句子,与泡泡玛特那些冗长、多俚语、中英混杂的评论不同(p. 5)。
Bernstein (Hu, M., Zhang, K., & Fong, F.). (2026, January 13). China consumer: Pop Mart China sales tracker: Deceleration in offline growth and productivity; risks in 2026 [Equity research report].
It gives independent store and online data showing Pop Mart's China demand slowing in late 2025.
The tracker covers about 70% of China revenue (p. 1). Data runs to November 2025.这份追踪覆盖约 70% 的中国收入(第 1 页)。数据截至 2025 年 11 月。
Revenue per store fell to RMB 1.07mn in November 2025, from RMB 1.30mn in October (p. 1).单店收入由 2025 年 10 月的 RMB 1.30mn(130 万元)降至 11 月的 RMB 1.07mn(107 万元)(第 1 页)。
Sales per square metre dropped to RMB 6,900 in November (p. 1). It had held near RMB 8,000 in August to October.11 月每平方米销售额降至 6,900 元(第 1 页)。8 月至 10 月一直维持在约 8,000 元。
Bernstein reads the drop as a sign that the existing collector base is tapped out (p. 5).伯恩斯坦(Bernstein)认为,这一下滑说明现有收藏者群体已接近饱和(第 5 页)。
Online GMV rose 144% YoY in November (p. 7). Bernstein links this to the 11.11 festival, not to new demand.11 月线上商品交易总额(GMV)同比增长 144%(第 7 页)。伯恩斯坦认为这是「双十一」(11.11)带动的,而非新增需求。
About 80% of online GMV came from Pop Mart flagship stores and about 20% from third-party sellers (p. 7).约 80% 的线上 GMV 来自泡泡玛特旗舰店,约 20% 来自第三方卖家(第 7 页)。
In 1H25, offline made up about 61% of China sales (p. 4). Pop Draw led China online sales with 38%.2025 年上半年(1H25),线下约占中国销售的 61%(第 4 页)。抽盒机(Pop Draw)以 38% 领跑中国线上销售。
Bernstein says growth now needs a steady run of blockbuster launches (p. 5). It calls this a hit-driven model.伯恩斯坦指出,增长如今要靠接连不断的爆款(第 5 页)。它称之为依赖爆款的模式(hit-driven model)。
For our report · 对报告的意义
Ch 4.1: Pop Draw is the largest single online channel in China, at 38% in 1H25 (p. 4).第 4.1 章:抽盒机是中国最大的单一线上渠道,2025 年上半年(1H25)占 38%(第 4 页)。
Ch 4.2: Online sales spike around 11.11 (p. 7). Part of that may be a shift from stores, not new buyers.第 4.2 章:线上销售在「双十一」(11.11)前后激增(第 7 页)。其中部分可能是从门店转移过来的,而非新顾客。
Ch 4.4: Store productivity fell, with sales per square metre down to RMB 6,900 in November (p. 1).第 4.4 章:门店效率下降,11 月每平方米销售额降至 6,900 元(第 1 页)。
Ch 10: Demand depends on hits (p. 5). When hits fade, the price premium is harder to hold.第 10 章:需求依赖爆款(第 5 页)。爆款退潮时,价格溢价就更难守住。
Limits · 局限
This is a sell-side view with an Underperform rating. It leans to bearish readings.这是卖方观点,评级为「跑输大市」(Underperform),解读偏向看淡。
Same-store figures come from a third-party panel, Meritco, not from company filings. The data stops at November 2025.同店数据来自第三方样本 Meritco,而非公司披露。数据截至 2025 年 11 月。
It does not measure resale prices or the secondary market.报告没有衡量转售价格或二手市场。
17. Uniting the Tribes: Using Text for Marketing Insight · 联合各个“部落”:用文本获取营销洞察
Conceptual review and methods guide (no new data) 概念综述与方法指南(无新数据)Ch 2, 8, 13
Berger, J., Humphreys, A., Ludwig, S., Moe, W. W., Netzer, O., & Schweidel, D. A. (2020). Uniting the tribes: Using text for marketing insight. Journal of Marketing, 84(1), 1–25. https://doi.org/10.1177/0022242919873106
Text both reflects the person who wrote it and impacts the person who reads it (p. 2).
Text can serve prediction (best accuracy) or understanding (why something happens) (pp. 5–6).文本可用于预测(追求准确度),也可用于理解(解释为什么发生)(pp. 5–6)。
The workflow has 4 steps: preprocess, extract, measure, then check validity (p. 7).分析流程分 4 步:预处理、抽取、量化,最后检验效度(p. 7)。
Validity is the main threat; dictionaries must be tested on the specific data, for example a 10% sample reaching an 80% hit rate (pp. 13–14).效度是最大的风险;词典必须在具体数据上检验,例如抽取 10% 样本,命中率达到 80%(pp. 13–14)。
Reviews, service calls, ads and reports create large amounts of text data (p. 1).在线评论、客服电话、广告和年报产生了大量文本数据(p. 1)。
By some estimates, 80%–95% of business data is unstructured, mostly text (p. 1).据一些估计,80%–95% 的商业数据是非结构化的,其中大部分是文本(p. 1)。
Automated text analysis is computer-assisted and frees researchers from fixed scales and scripted questions (p. 1).自动文本分析(automated text analysis)由计算机辅助,让研究者摆脱固定量表和预设问题的束缚(p. 1)。
The paper makes four contributions: prediction and understanding, a how-to guide, research propositions, and uniting the field (p. 2).本文有四个贡献:说明预测与理解、提供操作指南、提出研究命题,以及联合整个学科(p. 2)。
Marketing is split into "tribes", such as modelers, consumer psychologists, strategy and culture researchers (p. 2).营销学被分成多个“部落”,如建模者、消费心理学者、战略和文化研究者(p. 2)。
For our report · 对报告的意义
Ch 2: We describe our VoC pipeline with the 4 steps on p. 7: preprocess, extract, measure, validate.第 2 章:我们按 p. 7 的 4 个步骤描述客户之声流程:预处理、抽取、量化、检验。
Ch 8: We use a dictionary for clear signals (price, fake, hidden figure 隐藏款) and model-based coding for complex themes like regret (p. 11).第 8 章:清晰信号(价格、假货、隐藏款)用词典,复杂主题如后悔情绪用模型编码(p. 11)。
Ch 8: We hand-code a 10% sample and report hit rate, precision and recall, following Table 5 (pp. 13–14).第 8 章:我们按表 5 人工编码 10% 样本,并报告命中率、精确率和召回率(pp. 13–14)。
Ch 8: We avoid relying on topic models for short Douyin 抖音 live comments, since short slang text weakens them (p. 10).第 8 章:抖音直播评论短且多俚语,会削弱主题模型,因此我们不单靠主题模型(p. 10)。
Limits · 局限
It is a review; none of its advice is tested on new data here.这是一篇综述,文中建议没有在新数据上检验。
It predates large language models, so it gives no guidance on prompt-based coding.它早于大语言模型(LLM),因此没有关于提示词编码的指导。
Examples are mostly English and Western; Chinese word segmentation and internet slang are not covered.例子多为英文和西方语境;没有涉及中文分词和网络用语。
18. Enhancing User Engagement: The Role of Gamification in Mobile Apps · 提升用户投入:游戏化在移动应用中的作用
Empirical survey study with a structural equation model (PLS-SEM) 实证问卷研究,采用结构方程模型(PLS-SEM)Ch 6.2, 6.4, 7
Bitrián, P., Buil, I., & Catalán, S. (2021). Enhancing user engagement: The role of gamification in mobile apps. Journal of Business Research, 132, 170–185. https://doi.org/10.1016/j.jbusres.2021.04.028
A survey of 276 U.S. Fitbit app users tests how 3 groups of game elements build engagement (pp. 170, 176).
Achievement elements such as badges and progress bars meet all three needs: competence, autonomy and relatedness (p. 179).徽章、进度条等成就类元素同时满足三种需求:胜任感、自主感和归属感(p. 179)。
Social elements raise relatedness but cut autonomy (β = −0.456), so their total effect on engagement is negative (p. 179).社交类元素提升归属感,却削弱自主感(β = −0.456),因此对用户投入的总效应为负(p. 179)。
Engagement strongly predicts continued use (β = 0.738), word of mouth (β = 0.776) and app rating (β = 0.585) (p. 179).用户投入强烈预测持续使用(β = 0.738)、口碑传播(β = 0.776)和应用评分(β = 0.585)(p. 179)。
Global app downloads grew from 140.7 billion in 2016 to 204 billion in 2019 (p. 170).全球应用下载量从 2016 年的 1,407 亿次(140.7 billion)增至 2019 年的 2,040 亿次(204 billion)(p. 170)。
Yet only 32% of users open any one app more than 10 times (p. 170).但只有 32% 的用户会打开同一个应用超过 10 次(p. 170)。
About 25% of apps are used only once after download, so engagement is weak (p. 170).约 25% 的应用下载后只用一次,可见用户投入很弱(p. 170)。
Gamification means adding game-like features to a service to support gameful experience and value (p. 170).游戏化(gamification)指在服务中加入类似游戏的功能,以带来游戏般的体验和价值(p. 170)。
Gaps: few studies link game elements to theory, and most treat engagement as the end point (p. 170).研究缺口:很少研究把游戏元素与理论联系起来,且多数只把用户投入当作终点(p. 170)。
For our report · 对报告的意义
Ch 6.2: Pop Mart's collection progress and member points act like achievement elements, the group that meets all three needs (p. 179).第 6.2 章:泡泡玛特的收集进度和会员积分相当于成就类元素,而这类元素能满足全部三种需求(p. 179)。
Ch 6.2: Share-to-draw or group-buy prompts are social elements; this paper predicts they may cut autonomy (β = −0.456) (p. 179).第 6.2 章:分享抽盒或拼团提示属于社交类元素;本文预测它们可能削弱自主感(β = −0.456)(p. 179)。
Ch 6.4: Felt pressure from friends is a tested mechanism for the dark-pattern risk of social nudges (p. 180).第 6.4 章:来自朋友的压力感,是社交助推(nudge)带来暗黑模式风险的一个经过检验的机制(p. 180)。
Ch 7: We can audit the Pop Mart app and Pop Draw 抽盒机 against the 12 elements and score each one (p. 176).第 7 章:我们可以对照这 12 个元素审计泡泡玛特应用和抽盒机,并逐项打分(p. 176)。
Limits · 局限
One fitness app with U.S. MTurk users; no paid draw and no chance element like a blind box.只研究了一个健身应用和美国 MTurk 用户;没有付费抽取,也没有盲盒那样的随机元素。
Cross-sectional self-report data cannot prove cause and effect (p. 182).横截面自评数据无法证明因果关系(p. 182)。
Fitbit's social features are step contests; Pop Mart's social features are trading and showing off, which may work differently.Fitbit 的社交功能是步数比赛;泡泡玛特的社交功能是交换和炫耀,作用方式可能不同。
19. Does Gamification Affect Brand Engagement and Equity? A Study in Online Brand Communities · 游戏化会影响品牌投入和品牌资产吗?一项在线品牌社群研究
Empirical survey study with a structural equation model (PLS-SEM) and multi-group test 实证问卷研究,采用结构方程模型(PLS-SEM)和多组比较Ch 5, 6.2, 10
Xi, N., & Hamari, J. (2020). Does gamification affect brand engagement and equity? A study in online brand communities. Journal of Business Research, 109, 449–460. https://doi.org/10.1016/j.jbusres.2019.11.058
A survey of 824 members of the Xiaomi and Huawei brand communities in China links game features to brand engagement and brand equity (p. 449).
Achievement and social features link positively to all three kinds of brand engagement: emotional, cognitive and social (p. 455).成就类和社交类功能与三种品牌投入都正相关:情感投入、认知投入和社交投入(p. 455)。
Immersion features link only to social engagement (β = 0.084), and engagement links to awareness and mostly to loyalty (p. 455).沉浸类功能只与社交投入相关(β = 0.084);品牌投入与品牌知名度相关,也大多与忠诚度相关(p. 455)。
Effects are real but small: the model explains only 9.6% to 17.9% of variance (pp. 454–455).效应确实存在但较小:模型只解释了 9.6% 到 17.9% 的方差(pp. 454–455)。
Firms hope gamification will raise brand engagement, awareness and loyalty, but evidence is thin (p. 449).企业希望游戏化能提升品牌投入、知名度和忠诚度,但证据很少(p. 449)。
Loyalty schemes built on money or material rewards may build less long-term loyalty than intrinsic need satisfaction (p. 449).基于金钱或物质奖励的忠诚计划,在培养长期忠诚方面可能不如满足内在需求(p. 449)。
Many gamified business projects have failed after large investments, so firms are split on its value (p. 449).许多游戏化商业项目在大笔投入后失败,因此企业对其价值看法两极(p. 449)。
Earlier studies link gamification to brand attitude, awareness, engagement and brand love (p. 449).已有研究把游戏化与品牌态度、知名度、投入和品牌热爱联系起来(p. 449)。
Brand equity has not yet been studied as an outcome of gamification (p. 450).品牌资产(brand equity)尚未被当作游戏化的结果来研究(p. 450)。
For our report · 对报告的意义
Ch 6.2: Pop Mart member points, levels and Pop Draw 抽盒机 collection progress are achievement features, which link to all three engagement types (p. 455).第 6.2 章:泡泡玛特会员积分、等级和抽盒机收集进度属于成就类功能,与三种投入都相关(p. 455)。
Ch 6.2: We keep the claim modest, since features explain only 9.6% to 17.9% of variance (pp. 454–455).第 6.2 章:我们保持结论克制,因为功能只解释了 9.6% 到 17.9% 的方差(pp. 454–455)。
Ch 5: Showing off hidden figures and trading in community spaces maps onto social features, the largest path (β = 0.216) (p. 455).第 5 章:在社群中晒隐藏款和交换,对应社交类功能,这是最大的路径(β = 0.216)(p. 455)。
Ch 10: Loyalty includes willingness to pay a higher price, so engagement may help Pop Mart hold retail price (p. 454).第 10 章:忠诚度包含愿意支付更高价格,因此品牌投入可能帮助泡泡玛特守住零售价(p. 454)。
Limits · 局限
Two Chinese tech brands in 2017–2018; collectible toys with chance draws may behave differently.研究对象是 2017–2018 年的两个中国科技品牌;带随机抽取的收藏玩具可能表现不同。
Cross-sectional self-report data show association, not cause.横截面自评数据只能显示相关,不能证明因果。
The sample is community members, so people who never joined are missing.样本只包括社群成员,从未加入社群的人没有被纳入。
20. Resale or agency sale? Equilibrium analysis on the role of live streaming selling · 自营转售还是平台代销?直播带货作用的均衡分析
Analytical journal article (game-theory supply-chain model with numerical examples; no survey or sales data) 分析型期刊论文(博弈论供应链模型 + 数值算例;没有问卷或销售数据)Ch 4, 5, 10, 11
Hao, C., & Yang, L. (2023). Resale or agency sale? Equilibrium analysis on the role of live streaming selling. European Journal of Operational Research, 307, 1117–1134. https://doi.org/10.1016/j.ejor.2022.09.021
Careful: "resale" here means the platform buys stock and resells it itself, not secondhand trading.
The model has one brand (the supplier) and one platform, and both sell through a live stream and a regular store page.模型里只有一个品牌方(供应商)和一个平台,两者都通过直播渠道和普通商品页销售。
Without returns, agency sale (platform takes a commission) is the only format both sides agree on.不考虑退货时,代销(agency sale,平台抽佣)是双方唯一都能接受的模式。
Once high live-stream returns are added, resale can also suit both sides, mainly for price-controlled brands.一旦加入直播的高退货率,转售也可能让双方都满意,主要适用于严格控价的品牌。
Across almost all cases, a discount only in the live stream (strategy D) hurts the brand.几乎在所有情形下,只在直播间打折(策略 D)都会损害品牌方。
The paper builds live-stream selling models with two sales formats and three pricing strategies (p. 1117).论文建立直播销售模型,包含两种销售模式和三种定价策略(第 1117 页)。
The formats are resale (R) and agency sale (A). The strategies are same high (HH), same low (LL) and differential (D) (p. 1117).两种模式是转售(R)和代销(A);三种策略是同高价(HH)、同低价(LL)和差别定价(D)(第 1117 页)。
Common wisdom says agency sale reduces double marginalization and benefits both firms (p. 1117).传统观点认为,代销能减轻双重加价(double marginalization),对双方都有利(第 1117 页)。
With consumer returns, resale can become the better choice for both firms under HH and D (p. 1117).考虑消费者退货后,在 HH 和 D 策略下,转售可能对双方都更好(第 1117 页)。
Under HH the key driver is the commission rate; under D it is low-value buyers' valuation (p. 1117).HH 策略下的关键因素是佣金率;D 策略下是低估值买家的估值(第 1117 页)。
For our report · 对报告的意义
Ch 10: The paper finds a live-stream-only discount hurts the brand in almost all cases (p. 1124). This supports Pop Mart holding one retail price.第 10 章:论文发现只在直播间打折几乎总会损害品牌(第 1124 页),这支持泡泡玛特维持统一零售价。
Ch 10: Pop Mart's fixed price across channels looks like strategy HH. This mapping is ours, not the paper's.第 10 章:泡泡玛特各渠道统一定价,类似 HH 策略。这一对应是我们的判断,不是论文原话。
Ch 4: The paper treats self-run versus marketplace selling as a trade-off set by commission, returns and live-stream cost (p. 1118).第 4 章:论文把自营与平台入驻看作取舍,取决于佣金、退货和直播成本(第 1118 页)。
Ch 5: Use the return figures with care: 30% for top streamers and up to 70% for others (p. 1117).第 5 章:可谨慎引用退货数据:头部主播 30%,其他主播最高 70%(第 1117 页)。
Limits · 局限
It is a stylised theory model with no empirical test, so it shows directions, not sizes.这是简化的理论模型,没有实证检验,只能说明方向,不能说明大小。
Prices are fixed by strategy rather than chosen, which narrows what the model can say about pricing.价格由策略事先固定,而不是由企业选择,限制了模型对定价的解释力。
It assumes full refunds and rational buyers, but blind boxes 盲盒 are usually non-returnable and bought on impulse.它假设全额退款和理性消费者,但盲盒通常不退换,而且多为冲动购买。
21. Consumer Engagement in Social Media Brand Communities: A Literature Review · 社交媒体品牌社群中的消费者参与:文献综述
Systematic (descriptive) literature review of 134 journal articles, ending in 10 propositions and an integrative framework 系统性(描述型)文献综述,分析 134 篇期刊论文,最后提出 10 条命题和一个整合框架Ch 4, 5, 7, 8, 9
Santos, Z. R., Cheung, C. M. K., Coelho, P. S., & Rita, P. (2022). Consumer engagement in social media brand communities: A literature review. International Journal of Information Management, 63, Article 102457. https://doi.org/10.1016/j.ijinfomgt.2021.102457
The authors reviewed 134 articles on consumer engagement in social media brand communities (p. 1).
Research defines engagement in three ways: motivation to interact, attitude towards the community, and behaviour in it (p. 7).研究对参与有三种定义:互动动机、对社群的态度、在社群中的行为(p. 7)。
Engagement is driven by community, brand, social and mass-communication factors. It leads to community outcomes and brand outcomes such as loyalty and word of mouth (pp. 12–14).参与由社群、品牌、社会和大众传播四类因素驱动。它带来社群结果,也带来忠诚度、口碑等品牌结果(pp. 12–14)。
The paper closes with 10 propositions and one integrative framework (Fig. 3) as an agenda for future research (pp. 12–14).论文最后提出 10 条命题和一个整合框架(图 3),作为未来研究议程(pp. 12–14)。
Marketers use social media brand communities to connect consumers and help them share product and brand information (p. 1).营销者利用社交媒体品牌社群连接消费者,帮助他们交流产品与品牌信息(p. 1)。
Engagement in these communities helps build brand trust, brand loyalty and brand awareness (p. 1).在这些社群中的参与有助建立品牌信任、品牌忠诚和品牌知名度(p. 1)。
The main problem is the lack of a consistent definition. Engagement appears with many foci: brand, organisation, community and advertising (p. 1).主要问题是缺乏一致的定义。「参与」出现在品牌、组织、社群、广告等多种对象上(p. 1)。
Brodie et al. (2011) is seen as the pioneer study that brought engagement into marketing (p. 2).Brodie 等人(2011)被视为把「参与」概念引入营销学的开创性研究(p. 2)。
Three research streams exist: engagement as a psychological state, as intrinsic motivation, and as behaviour (p. 2).现有三条研究脉络:把参与视为心理状态、内在动机或行为(p. 2)。
For our report · 对报告的意义
Ch 4: Pop Mart 泡泡玛特 runs brand-managed communities, while fans run many others. The paper says fan-run ones feel more trustworthy (p. 12).第 4 章:泡泡玛特运营品牌官方社群,而许多社群由粉丝自发运营。论文指出粉丝社群更可信(p. 12)。
Ch 5: Use the four antecedent groups (community, brand, social, mass communication) to organise drivers of engagement on Xiaohongshu 小红书 and Douyin 抖音 (pp. 12–13).第 5 章:用四类前因(社群、品牌、社会、大众传播)整理小红书和抖音上的参与驱动因素(pp. 12–13)。
Ch 5: Social factors fit Labubu well. Celebrity links and a prestigious image attract people who want to belong (p. 13).第 5 章:社会因素很适合解释 Labubu。名人关联和高端形象会吸引想要归属的人(p. 13)。
Ch 8: Brand communities work as a research tool in a natural setting. This supports our Voice of the Customer corpus of community posts (p. 15).第 8 章:品牌社群可在自然环境中充当研究工具。这支持我们用社群帖子建立顾客之声语料库(p. 15)。
Limits · 局限
Game communities were excluded on purpose, so the review says little about gamified or draw-based engagement such as Pop Draw 抽盒机 (p. 4).游戏社群被刻意排除,因此该综述对游戏化或抽取式参与(如 Pop Draw 抽盒机)谈得很少(p. 4)。
The evidence base is mostly Facebook and Twitter. Only Weibo is named among Chinese platforms, so transfer to Xiaohongshu 小红书 and Douyin 抖音 is untested (p. 6).证据主要来自 Facebook 和 Twitter,中国平台只提到微博,因此能否迁移到小红书和抖音尚未检验(p. 6)。
Screening was done by the first author alone. This raises the risk of selection bias (p. 4).论文筛选由第一作者独自完成,增加了选择偏差的风险(p. 4)。
22. Impact of Product Characteristics of Limited Edition Shoes on Perceived Value, Brand Trust, and Purchase Intention; Focused on the Scarcity Message Frequency · 限量版球鞋的产品特征对感知价值、品牌信任和购买意向的影响:聚焦稀缺信息的频率
Online survey with a two-group design (one-time vs repeated limited release), analysed by structural equation modelling and multi-group comparison 线上问卷调查,两组设计(一次性限量与重复限量发售),用结构方程模型和多群组比较分析Ch 6.3, 9, 10
Chae, H., Kim, S., Lee, J., & Park, K. (2020). Impact of product characteristics of limited edition shoes on perceived value, brand trust, and purchase intention; focused on the scarcity message frequency. Journal of Business Research, 120, 398–406. https://doi.org/10.1016/j.jbusres.2019.11.040
The paper surveyed 460 young Korean buyers of limited-edition sneakers (p. 401).
Three product features mattered: uniqueness, self-expression and investment. Investment had the largest effect on value (pp. 401–403).三个产品特征起作用:独特性、自我表达和投资性。投资性对价值的影响最大(pp. 401–403)。
Economical value drove purchase intention most (β = 0.578). Brand trust added only a small effect (β = 0.069) (p. 403).经济价值对购买意向的推动最大(β = 0.578)。品牌信任只带来很小的影响(β = 0.069)(p. 403)。
Repeated releases did not weaken the model. They strengthened the links from investment to emotional and social value (p. 404).重复发售并没有削弱模型。它反而加强了投资性对情感价值和社会价值的影响(p. 404)。
Trading up means buying goods with higher emotional value to feel achievement (p. 398).消费升级(trading up)指购买情感价值更高的商品以获得成就感(p. 398)。
Social media gives people a place to show rare products, and acts as a reference group (p. 398).社交媒体为人们提供展示稀有商品的场所,也充当参照群体(p. 398)。
The sneaker market is expected to reach $95.14 billion by 2025. Nike and Adidas market heavily with limited editions (p. 398).球鞋市场预计到 2025 年将达 $95.14 billion(951.4 亿美元)。耐克和阿迪达斯大量使用限量版营销(p. 398)。
Past work studied limited-time and limited-quantity messages. Little research looks at the limited product itself (p. 398).以往研究关注限时和限量信息,很少研究限量产品本身(p. 398)。
Repeated scarcity messages can make buyers suspect a sales gimmick and lose purchase intention (pp. 398–399).重复的稀缺信息会让买家怀疑这是促销噱头,从而失去购买意向(pp. 398–399)。
For our report · 对报告的意义
Ch 6.3: Scarcity works partly through resale value. Investment had the largest effect on economical value (β = 0.823) (p. 403).第 6.3 章:稀缺部分通过转售价值起作用。投资性对经济价值的影响最大(β = 0.823)(p. 403)。
Ch 6.3: The backfire effect warns that constant limited-drop messages in the Pop Mart 泡泡玛特 app may lose force (p. 401).第 6.3 章:反效果提醒我们,泡泡玛特 App 中不断出现的限量发售信息可能失去作用(p. 401)。
Ch 10: Re-releases did not reduce value here. Predictable resale prices made repeated releases stronger on some paths (p. 404).第 10 章:本研究中复刻并没有降低价值。可预测的转售价格使重复发售在部分路径上更强(p. 404)。
Ch 10: Labubu restocks may shift buyers from speculation to trust. Trust mattered more under repeated release (0.149 versus 0.021) (p. 404).第 10 章:Labubu 补货可能使买家从投机转向信任。重复发售下信任作用更大(0.149 对 0.021)(p. 404)。
Limits · 局限
The sample is Korean sneaker buyers in 2018. Designer toys and blind boxes in China are a different product and market.样本是 2018 年的韩国球鞋买家。中国的潮玩和盲盒是不同的产品和市场。
The scarcity manipulation was one shoe image per group, with only one product for the one-time condition (p. 405).稀缺操纵只是每组看一张鞋图,一次性条件只用了 1 款产品(p. 405)。
Some statistics look inconsistent. Self-expression → economical value has C.R. 1.99 but is marked p < .001 (p. 403).部分统计数据看起来不一致。自我表达→经济价值的 C.R. 为 1.99,却标为 p < .001(p. 403)。
Summaries are written by the group from sources we read. Numbers and page references were checked against the originals. Licensed cases and reports are summarised, not reproduced.
摘要由小组根据所读文献整理,数字和页码已对照原文核对。授权案例与报告只作摘要,不转载原文。