6. UI Dark Patterns and Where to Find Them: A Study on Mobile Applications and User Perception
UI 暗黑模式在哪里:一项关于移动应用与用户感知的研究 · Empirical HCI conference paper: expert coding of 240 Android apps (Study 1) plus an online video survey of 589 users (Study 2) 实证型人机交互(HCI)会议论文:专家对 240 款安卓应用编码(研究一),加上 589 名用户的在线视频问卷(研究二) · Ch 6.3, 6.4, 7, 8, 13 · 30 min
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
The paper in 60 seconds · 60 秒读懂
The authors coded dark patterns in 240 trending free Android apps by using each app for ten minutes (pp. 3–4).作者把 240 款热门免费安卓应用各实际使用十分钟,再对其中的暗黑模式(dark pattern)编码(pp. 3–4)。
95% of apps had at least one dark pattern, with 7.4 per app on average (p. 5).95% 的应用至少含一种暗黑模式,平均每款 7.4 个(p. 5)。
Nagging, False Hierarchy and Preselection were the most common subtypes (p. 5).最常见的子类是唠叨打断(Nagging)、虚假层级(False Hierarchy)和预选(Preselection)(p. 5)。
In a survey of 589 people, 55% of answers missed the dark pattern and 20% were unsure (pp. 7–8).在 589 人的问卷中,55% 的回答没看出暗黑模式,20% 表示不确定(pp. 7–8)。
Why we read it · 为什么要读
It gives us a ready taxonomy and coding rules for the Pop Mart app and Pop Draw audit in Ch 7.它为第 7 章审计泡泡玛特 App 和 Pop Draw 抽盒机提供了现成的分类法和编码规则。
It shows that users rarely notice dark patterns, so complaints alone cannot prove a design is fair (p. 8).它说明用户很少察觉暗黑模式,所以只看投诉无法证明设计是公平的(p. 8)。
It labels countdowns, multiple currencies and forced rewards as dark patterns, which matter for Ch 6.3 and 6.4 (p. 3).它把倒计时、多种虚拟货币和强制领奖励都归为暗黑模式,这与第 6.3 和 6.4 章直接相关(p. 3)。
20-minute reading path · 20 分钟阅读路线
Pages 页码
What to read
读什么
3
The five-type taxonomy and its subtypes.
五大类分类法及其子类。
4
The recording tasks and coding rules.
录屏任务和编码规则。
5–6
Table 1, the prevalence results and Figure 1.
表 1、普遍性结果和图 1。
8–9
DP-blindness results and Table 2.
暗黑模式盲区结果和表 2。
10
Implications and limitations.
启示与局限。
Abstract and Introduction · 摘要与引言
Pages · 页码 1–2
A dark pattern (DP) is an interface built to trick users into actions they did not intend (p. 1).暗黑模式(dark pattern, DP)指故意设计来诱导用户做出本不想做的操作的界面(p. 1)。
DPs can also push users to over-share personal data or grant extra permissions (p. 1).暗黑模式还会让用户过度分享个人数据,或授予超出本意的权限(p. 1)。
Earlier work found that 11% of 11k shopping websites used DPs (Mathur et al.) (p. 1).此前 Mathur 等人发现,11k 个购物网站中有 11% 使用了暗黑模式(p. 1)。
The paper asks two new questions: how common DPs are in mobile apps, and whether users recognise them (p. 1).本文提出两个新问题:暗黑模式在移动应用中有多普遍,以及用户能否认出它们(p. 1)。
Unlike screenshot studies, two researchers actively used each app, like a cognitive walkthrough (p. 1).与只看截图的研究不同,两位研究者实际操作每款应用,类似认知走查(cognitive walkthrough)(p. 1)。
Background: Ethical UX design and neuromarketing · 背景:伦理化 UX 设计与神经营销
Pages · 页码 2
Neuromarketing research finds the fear of missing out strongly influences users (p. 2).神经营销(neuromarketing)研究发现,「怕错过」的感受对用户影响很大(p. 2)。
Shops therefore use countdowns and limited offers, which become unethical when used to coerce (p. 2).因此网店常用倒计时和限时优惠;一旦用于胁迫,就变得不合伦理(p. 2)。
Researchers classify artificial urgency and scarcity as a dark pattern (p. 2).研究者把人为制造的紧迫感和稀缺感归为暗黑模式(p. 2)。
95% of adolescents have smartphone access, and minors are easier to manipulate (p. 2).95% 的青少年能使用智能手机,而未成年人更容易被操纵(p. 2)。
Luguri and Strahilevitz found 26% (mild DP) and 42% (aggressive DP) accepted a paid plan, against 11% without DPs (p. 2).Luguri 与 Strahilevitz 发现,温和版暗黑模式下 26%、激进版下 42% 的人接受付费计划,无暗黑模式时只有 11%(p. 2)。
The authors propose DP-blindness: users may stop seeing mild DPs, like banner blindness (p. 2).作者提出「暗黑模式盲区」(DP-blindness):用户可能对温和的暗黑模式视而不见,类似广告横幅盲区(p. 2)。
Background: Taxonomies of dark patterns (Gray et al.) · 背景:暗黑模式分类法(Gray 等人)
Pages · 页码 2–3
Conti and Sobiesk proposed eleven classes with twenty subclasses, such as Distraction and Forced Work (pp. 2–3).Conti 与 Sobiesk 提出十一个大类、二十个子类,例如分散注意力和强迫劳动(pp. 2–3)。
The paper adopts Gray et al.'s five types because that taxonomy is the most recent (p. 3).本文采用 Gray 等人的五大类,因为它是最新的分类法(p. 3)。
Nagging: repeated redirection away from the current task, for example popups (p. 3).唠叨打断(Nagging):一次或多次把用户从当前任务中拉走,例如弹窗(p. 3)。
Obstruction blocks the task: Intermediate Currency, Price Comparison Prevention and Roach Motel (p. 3).阻碍(Obstruction)让任务更难完成,含中间货币(Intermediate Currency)、阻止比价(Price Comparison Prevention)和蟑螂屋(Roach Motel)(p. 3)。
Sneaking hides key information: Bait and Switch, Hidden Costs, Sneak into Basket, Forced Continuity (p. 3).偷偷摸摸(Sneaking)隐藏关键信息,含诱饵调包(Bait and Switch)、隐藏费用(Hidden Costs)、偷加购物车(Sneak into Basket)和强制续订(Forced Continuity)(p. 3)。
Interface Interference biases the UI: Hidden Information, Preselection and Aesthetic Manipulation (p. 3).界面干扰(Interface Interference)让界面偏向某一选项,含隐藏信息(Hidden Information)、预选(Preselection)和美学操纵(Aesthetic Manipulation)(p. 3)。
Aesthetic Manipulation has four parts: Toying with Emotions, False Hierarchy, Disguised Ad, Trick Questions (p. 3).美学操纵下再分四种:玩弄情绪(Toying with Emotions,如优惠倒计时)、虚假层级、伪装广告(Disguised Ad)和陷阱问题(Trick Questions)(p. 3)。
Forced Action coerces tasks for a reward: Social Pyramid, Privacy Zuckering, Gamification (p. 3).强制行动(Forced Action)逼用户做事才给好处,含社交金字塔(Social Pyramid)、隐私诱骗(Privacy Zuckering)和游戏化(Gamification,强迫刷任务)(p. 3)。
Study 1: Corpus generation · 研究一:样本构建
Pages · 页码 3–4
Apps had to be on Android, free to download and trending in the US market (p. 3).入选条件:安卓平台、免费下载、在美国市场处于热门趋势(p. 3)。
Eight categories were used; Personalization was swapped for News and Magazines (p. 3).共用八个类别;个性化(Personalization)类全是启动器,被替换为新闻与杂志类(p. 3)。
A crawler pulled the top 400 trending apps per category from SensorTower on 12–13 July (p. 3).爬虫于 7 月 12–13 日从 SensorTower 抓取每类前 400 款热门应用(p. 3)。
The top 30 per category were kept, skipping unavailable, duplicate or launcher apps, for 240 apps (p. 3).每类保留前 30 款,跳过已下架、重复或启动器类应用,合计 240 款(p. 3)。
Study 1: Recording and classification method · 研究一:录屏与编码方法
Pages · 页码 4
Apps were studied in use, because some DPs such as Bait and Switch only show through interaction (p. 4).作者在使用过程中观察应用,因为诱饵调包等暗黑模式只有在交互中才会显现(p. 4)。
Two authors each recorded half the apps on a One Plus 5, with new Gmail accounts and SIM cards (p. 4).两位作者各录一半应用,使用 One Plus 5 手机、新注册的 Gmail 账号和新 SIM 卡(p. 4)。
Each app was used for ten minutes, giving 2,400 minutes (40 hours) of video (p. 4).每款应用使用十分钟,共得 2,400 分钟(40 小时)录像(p. 4)。
Fixed tasks: create account and log out, reopen app, market page, settings, shop to checkout, select names, normal use (p. 4).固定任务:注册并登出、关闭重开、进入商城页、进入设置页、购物至结账、尝试选中商品名称、按用途正常使用(p. 4)。
No purchases were made; shopping stopped right before payment (p. 4).全程没有真实购买,购物流程停在付款前一步(p. 4)。
A pilot of 40 apps (five per category) was coded jointly by the first two authors (p. 4).先随机抽 40 款应用(每类五款)作试点,由前两位作者共同编码(p. 4)。
Disagreements went to a third expert and were settled by majority vote (p. 4).分歧交给第三位专家讨论,以多数票决定(p. 4)。
No inter-rater agreement statistic such as kappa is reported; coding was done in pairs instead (p. 4).论文没有报告评分者一致性系数(如 kappa),而是采用两人一起编码的方式(p. 4)。
The remaining 200 apps were also coded jointly, since DP-blindness affected experts too (p. 4).其余 200 款也一起编码,因为作者发现专家同样会出现暗黑模式盲区(p. 4)。
Classification took 120 hours, and the sheet was double-checked afterwards (p. 4).编码共耗时 120 小时,完成后再复核编码表(p. 4)。
Study 1: Taxonomy adaptation and coding rules · 研究一:分类法调整与编码规则
Pages · 页码 4–5
Forced Continuity was dropped because the team took no paid subscriptions (p. 4).强制续订被排除,因为团队没有订阅任何付费服务(p. 4)。
Gamification was dropped because forced grinding is hard to see in ten minutes (p. 4).游戏化被排除,因为强迫刷任务在头十分钟里很难看到(p. 4)。
Watching an ad to unlock a feature was added under Forced Action (p. 4).「看广告解锁功能」被新增归入强制行动(p. 4)。
Any UI that seemed to benefit the app over the user was coded as a DP, regardless of intent (p. 4).只要界面看来对应用比对用户更有利,就记为暗黑模式,不论设计者本意(p. 4)。
Bait and Switch was narrowed to free-looking features that turn out to be premium (p. 5).诱饵调包被收窄为:看似免费的功能点开后才发现要付费或另装应用(p. 5)。
The final scheme had 33 DP cases across 16 subcategories; twelve cases could count only once per app (p. 5).最终方案有 33 种暗黑模式情形,分属 16 个子类;其中十二种每款应用只能计一次(p. 5)。
Study 1: Results · 研究一:结果
Pages · 页码 5–6
95% of the 240 apps contained one or more DPs (p. 5).240 款应用中 95% 含一种或以上暗黑模式(p. 5)。
1,787 DPs were found, averaging 7.4 per app (standard deviation 5) (p. 5).共发现 1,787 个暗黑模式,平均每款 7.4 个(标准差 5)(p. 5)。
About 10% of apps had 0–2 DPs (N=33), 37% had 3–6 (N=89) and 49% had 7 or more (N=118) (p. 5).约 10% 的应用有 0–2 个(N=33),37% 有 3–6 个(N=89),49% 有 7 个或以上(N=118)(p. 5)。
Only twelve apps (5%) had none, including Snapseed, Lego Juniors and Barcode Scanner (p. 5).只有十二款(5%)完全没有,包括 Snapseed、Lego Juniors 和 Barcode Scanner(p. 5)。
Call Free had the most: 23 DPs across 10 subcategories; Wish followed with 20 in 8 (p. 5).Call Free 最多,有 23 个,横跨 10 个子类;Wish 其次,20 个,横跨 8 个子类(p. 5)。
Apps averaged 2.7 of the five macro-categories (std 1.1); 37% had 3 (N=89) (p. 5).每款应用平均涉及五大类中的 2.7 类(标准差 1.1);37% 涉及 3 类(N=89)(p. 5)。
25% had 4 or 5 macro-categories (N=62), 23% had 2 (N=55), 14% had 1 or none (N=34) (p. 5).25% 涉及 4 或 5 类(N=62),23% 涉及 2 类(N=55),14% 涉及 1 类或没有(N=34)(p. 5)。
Apps averaged 4.3 of the 16 subcategories (std 2.6); 63% (N=152) had at least four (p. 5).平均涉及 16 个子类中的 4.3 个(标准差 2.6);63%(N=152)至少有四个(p. 5)。
Nagging was most frequent: 55% of apps interrupted users for permissions, ratings or ads (p. 5).唠叨打断最常见:55% 的应用会为索取权限、求评分或展示广告而打断用户(p. 5)。
False Hierarchy appeared in 61% of apps, often by styling the app-friendly option more prominently (p. 5).虚假层级出现在 61% 的应用中,通常把对应用有利的选项做得更醒目(p. 5)。
Preselection appeared in 60%; pre-ticked notifications were most common (N=121), with 81 apps pre-ticking more than two (p. 5).预选出现在 60% 的应用中;最常见的是默认勾选通知(N=121),其中 81 款预选了两项以上(p. 5)。
31% of apps had Privacy Zuckering, such as tiny consent labels or default 'send usage data' (p. 6).31% 的应用存在隐私诱骗,例如同意条款字体极小,或默认开启「发送使用数据」(p. 6)。
Roach Motel (no logout or no account deletion) appeared in 41% of apps, including Spotify and Instagram (p. 6).蟑螂屋(无法登出或无法删除账号)出现在 41% 的应用中,包括 Spotify 和 Instagram(p. 6)。
Price Comparison Prevention was found 23 times, but in 77% of the 30 Shopping apps (p. 6).阻止比价共出现 23 次,但在 30 款购物类应用中占 77%(p. 6)。
Welch ANOVA showed News and Magazines had fewer DPs than five other categories (F[7,232] = 3.390, p < 0.05) (p. 6).韦尔奇方差分析(Welch ANOVA)显示,新闻与杂志类的暗黑模式少于其他五类(F[7,232] = 3.390,p < 0.05)(p. 6)。
Study 1: Discussion · 研究一:讨论
Pages · 页码 6
Most trending apps use some malicious design to steer user responses (p. 6).大多数热门应用都使用某种恶意设计来引导用户反应(p. 6)。
Study 2: Online experiment design · 研究二:在线实验设计
Pages · 页码 7
The survey was framed as rating app user experience, and $2 per person went to charity (p. 7).问卷以「评价应用体验」为名,每位参与者可为其选定的慈善机构捐出 2 美元(p. 7)。
Each person saw three apps: two with DPs, rotated from five, then Lego Juniors as a DP-free control (p. 7).每人看三款应用:两款含暗黑模式(从五款中轮换抽取),最后一款是无暗黑模式的对照组 Lego Juniors(p. 7)。
The five DPs were Nagging, Intermediate Currency, False Hierarchy, Forced Action and Sneak into Basket (p. 7).五种暗黑模式为唠叨打断、中间货币、虚假层级、强制行动和偷加购物车(p. 7)。
Apps: Face Reading, Romwe, Tag with Ryan, Talkatone and Roblox (p. 7).对应应用:Face Reading、Romwe、Tag with Ryan、Talkatone 和 Roblox(p. 7)。
Users watched a 30-second video and rated ease of use on a 5-point Likert scale (p. 7).参与者观看 30 秒视频,并用 5 点李克特量表(Likert scale)评价易用性(p. 7)。
On a new page they were asked if they saw any 'malicious design'; the term DP was never used (p. 7).之后在新页面上问他们是否看到「恶意设计」;全程不用「暗黑模式」一词,避免先入为主(p. 7)。
Sneak into Basket could only be shown together with Preselection, a noted confound (p. 7).偷加购物车只能与预选同时出现,作者把这一混杂因素记录在案(p. 7)。
Ages ran from 19 to 77 (mean 30.3, SD 10.75), from 46 countries (p. 8).年龄 19 至 77 岁(平均 30.3,标准差 10.75),来自 46 个国家(p. 8)。
98% used smartphones daily, and 88% had no prior familiarity with the test apps (p. 8).98% 每天使用智能手机,88% 从未接触过测试应用(p. 8)。
Study 2: Results on DP-blindness · 研究二:暗黑模式盲区结果
Pages · 页码 8–9
1,767 answers were collected (1,178 excluding the control) (p. 8).共收集 1,767 个回答(不含对照组为 1,178 个)(p. 8)。
For DP apps, 55% said no malicious design, 20% were unsure and 25% said yes (p. 8).对含暗黑模式的应用,55% 回答没有恶意设计,20% 不确定,25% 回答有(p. 8)。
For the control app, 86% correctly said it had no DPs (p. 8).对照组应用中,86% 正确判断没有暗黑模式(p. 8)。
Only 7% of 366 open comments on ratings mentioned a DP before priming (p. 8).在提示前写评分理由的 366 人中,只有 7% 提到暗黑模式(p. 8)。
Of those who spotted something, only 24% matched the actual DP; 56% were unsure or named another (p. 8).自认发现问题的人中,只有 24% 与真正的暗黑模式一致;56% 不确定或指的是别的设计(p. 8)。
Detection differed by app: chi-square(10) = 221.167, p < 0.001 (p. 8).不同应用的识别率差异显著:卡方(10)= 221.167,p < 0.001(p. 8)。
Romwe's pre-added insurance was spotted least, by 14%, significantly worse than all others (p < 0.05) (p. 8).Romwe 默认加购保险最难被发现,识别率仅 14%,显著低于其他所有应用(p < 0.05)(p. 8)。
Users got more attentive after the first app: chi-square(4) = 58.201, p < 0.001 (p. 9).看完第一款后用户更警觉:卡方(4)= 58.201,p < 0.001(p. 9)。
Prior experience was linked to detection (chi-square(6) = 81.699, p < 0.001); age and education were not (p. 9).既往经验与识别率相关(卡方(6)= 81.699,p < 0.001);年龄、就业和学历无关(p. 9)。
Study 2: Learnability and behaviour change · 研究二:学习效果与行为改变
Pages · 页码 9
49% said they learned something from the study, 30% somewhat and 12% no (p. 9).49% 表示从实验中学到东西,30% 表示有一点,12% 表示没有(p. 9)。
32% would change how they download apps extremely or moderately, 44% slightly, 24% not at all (p. 9).32% 会大幅或中度改变下载应用的方式,44% 只会轻微改变,24% 完全不变(p. 9)。
Spotting a DP was not linked to download intent (chi-square(8) = 5.147, p = 0.742) (p. 9).是否发现暗黑模式与下载意愿无关(卡方(8)= 5.147,p = 0.742)(p. 9)。
It was linked to use, rating and recommending: chi-square(8) = 15.680, 19.073 and 21.038, all p < 0.05 (p. 9).但与使用、评分和推荐给朋友相关:卡方(8)分别为 15.680、19.073 和 21.038,均 p < 0.05(p. 9)。
Participants raised concern about children being targeted by frequent ads (p. 9).参与者担心儿童被频繁的广告针对(p. 9)。
Implications of the studies · 研究启示
Pages · 页码 9–10
More empirical work is needed on why designers add DPs and which types harm most (p. 10).需要更多实证研究,弄清设计者为何加入暗黑模式,以及哪类危害最大(p. 10)。
Automated tools are needed to detect DPs and alert users (p. 10).需要自动化工具来检测暗黑模式并提醒用户(p. 10)。
Users are told to pre-screen apps and decide carefully what data to share (p. 10).作者建议用户预先检查应用,并谨慎决定分享哪些数据(p. 10)。
Limitations · 局限
Pages · 页码 10
Some definitions had to be interpreted, and borderline cases depend on the viewer (p. 10).部分定义需要自行解读,边界案例的判断因人而异(p. 10)。
Only free Android apps and the first ten minutes were covered; paid apps and later DPs may differ (p. 10).只覆盖免费安卓应用和使用头十分钟;付费应用和后期出现的暗黑模式可能不同(p. 10)。
Watching a 30-second video is not the same as using the app, so detection may be lower (p. 10).看 30 秒视频不同于亲手使用,识别率可能因此偏低(p. 10)。
Sneak into Basket was mixed with Preselection, though comments focused on the basket item (p. 10).偷加购物车与预选混在一起,但参与者评论主要针对被加入的商品(p. 10)。
Conclusion and future work · 结论与未来工作
Pages · 页码 11
The conclusion restates 240 apps and 95% prevalence, but cites 584 respondents rather than 589 (p. 11).结论重申 240 款应用、95% 的普遍率,但受访人数写成 584 而非 589(p. 11)。
Future work aims at methods to identify, characterise and raise awareness of DPs (p. 11).未来工作聚焦识别、刻画暗黑模式并提高用户意识的方法(p. 11)。
Figures and tables · 图表解读
Exhibit 图表
How to read it
怎么看
Table 1 (p. 5)
Lists the 33 DP cases, each mapped to subclass codes and marked S (once per app) or M (many). Read it as the coding manual.
Bars show total DPs per subcategory; dots show the share of apps with each. Nagging, False Hierarchy and Preselection top the chart.
柱形表示各子类的暗黑模式总数,点表示含该子类的应用比例。唠叨打断、虚假层级和预选位居前列。
Figure 2 (p. 6)
A Daily Pixel rating popup that both interrupts the user and visually favours 5 stars.
Daily Pixel 应用的评分弹窗,既打断用户,又在视觉上偏向 5 星。
Figure 3 (p. 7)
Survey flow: intro, then two DP apps and one control app, each with familiarity, video, rating and 'malicious design?' steps, then screenshots and background.
Counts of how much learning about DPs would change downloading, use, rating and recommending. Some columns look garbled in the text extraction, so check the PDF.
An interface designed to trick users into actions they did not intend.
故意诱导用户做出本不想做的操作的界面设计。
DP-blindness 暗黑模式盲区
Users failing to notice dark patterns, much like ignoring banner ads.
用户对暗黑模式视而不见,类似对横幅广告的忽视。
Nagging 唠叨打断
Repeatedly pulling the user away from the task, often with popups.
反复用弹窗等方式打断用户当前任务。
Intermediate Currency 中间货币
In-app tokens or gems that hide the real money cost.
用应用内代币或宝石掩盖真实花费。
Roach Motel 蟑螂屋
Easy to get in, such as signing up, but hard to get out, such as deleting an account.
进去容易(注册),出来难(删除账号)。
Sneak into Basket 偷加购物车
Adding unwanted items, such as insurance, to the cart by default.
默认把保险等用户不要的商品加入购物车。
Preselection 预选
Options that favour the app are ticked by default.
对应用有利的选项被默认勾选。
False Hierarchy 虚假层级
The app-friendly option is made bigger or brighter than the alternative.
把对应用有利的选项做得更大、更醒目。
Toying with Emotions 玩弄情绪
Using emotion, such as countdowns on offers or shaming, to push a choice.
用优惠倒计时、羞辱用户等情绪手段推动选择。
Forced Action 强制行动
Making users do a task, like watching an ad or logging in, to get something.
要求用户完成某任务(如看广告、登录)才能获得好处。
Gamification (as a DP) 游戏化(作为暗黑模式)
Forcing dull repeated tasks to earn what could otherwise be bought.
强迫用户重复枯燥任务,来换取本可用钱买到的东西。
Privacy Zuckering 隐私诱骗
Tricking users into sharing more personal data than they meant to.
诱导用户分享超出本意的个人数据。
Numbers we can cite · 可引用的数字
Value 数值
What it measures
含义
Page 页
240
Trending free Android apps coded in Study 1
研究 1(研究一)编码的热门免费安卓应用数量
p. 3
2,400
Minutes of recorded app use (ten minutes per app)
录制的应用使用分钟数(每款十分钟)
p. 4
120
Hours spent on classification
编码耗费的小时数
p. 4
95%
Share of apps with at least one dark pattern
至少含一种暗黑模式的应用比例
p. 5
1,787
Total dark patterns found
发现的暗黑模式总数
p. 5
7.4
Average dark patterns per app
每款应用平均暗黑模式数
p. 5
4.3
Average subcategories per app out of 16
每款应用平均涉及的子类数(共 16 个)
p. 5
61%
Apps with False Hierarchy
含虚假层级的应用比例
p. 5
60%
Apps with Preselection
含预选的应用比例
p. 5
55%
Apps that interrupted users (Nagging)
会打断用户(唠叨打断)的应用比例
p. 5
23
DPs in Call Free, the worst app
表现最差的 Call Free 所含暗黑模式数
p. 5
31%
Apps with Privacy Zuckering
含隐私诱骗的应用比例
p. 6
41%
Apps with Roach Motel (no logout or no account deletion)
含蟑螂屋(无法登出或删除账号)的应用比例
p. 6
77%
Shopping apps with Price Comparison Prevention
含阻止比价的购物类应用比例
p. 6
589
Completed survey responses in Study 2
研究 2(研究二)的完整问卷数
p. 7
46
Countries of survey participants
问卷参与者来自的国家数
p. 8
55%
Answers on DP apps that saw no malicious design
含暗黑模式应用中回答「没看到恶意设计」的比例
p. 8
14%
Detection rate for Romwe's pre-added insurance
Romwe 默认加购保险的识别率
p. 8
86%
Control answers correctly saying no DPs
对照组中正确判断无暗黑模式的比例
p. 8
24%
Spotters whose answer matched the real DP
自认发现者中答案与真实暗黑模式一致的比例
p. 8
221.167
Chi-square for detection differences across apps
不同应用识别差异的卡方值
p. 8
58.201
Chi-square showing more attention after the first app
显示看完第一款后更警觉的卡方值
p. 9
32%
Would change app downloading extremely or moderately
会大幅或中度改变下载方式的比例
p. 9
42%
Acceptance of a paid plan under an aggressive DP in Luguri and Strahilevitz
Luguri 与 Strahilevitz 研究中激进暗黑模式下接受付费计划的比例
p. 2
What it means for our report · 对报告的意义
Ch 6.4: Pop Draw's countdowns, coins and hint items map onto Toying with Emotions, Intermediate Currency and Gamification (p. 3).第 6.4 章:Pop Draw 抽盒机的倒计时、金币和提示道具,可分别对应玩弄情绪、中间货币和游戏化(p. 3)。
Ch 6.4: Most users miss dark patterns (55% no, 20% unsure), so low complaint volume is weak evidence of fairness (p. 8).第 6.4 章:多数用户察觉不到暗黑模式(55% 没看到,20% 不确定),所以投诉少并不能有力证明设计公平(p. 8)。
Ch 7: Reuse the seven-task walkthrough script and the Table 1 codebook for the Pop Mart app and Tmall, Douyin stores (p. 4–5).第 7 章:可沿用其七项任务走查脚本和表 1 编码表,审计泡泡玛特 App 以及天猫 Tmall、抖音 Douyin 店铺(p. 4–5)。
Ch 7: Check checkout for pre-ticked add-ons, since Romwe's insurance was spotted by only 14% (p. 8).第 7 章:重点检查结账页是否有默认勾选的附加项,因为 Romwe 的默认保险只有 14% 的人发现(p. 8)。
Ch 6.3: Artificial urgency and scarcity count as a dark pattern, so our scarcity cues need a fairness test (p. 2).第 6.3 章:人为制造的紧迫和稀缺被视为暗黑模式,因此我们分析的稀缺提示需要做公平性检验(p. 2)。
Ch 13: Like the authors, we should code in pairs and log longer sessions, because Gamification needs more than ten minutes (p. 4).第 13 章:应像作者一样两人编码,并延长观察时长,因为游戏化在十分钟内难以看到(p. 4)。
Limits and where it does not fit · 局限与不适用之处
The sample is US Android apps only; Chinese apps, WeChat mini-programs and Pop Mart were not studied (p. 3).样本仅限美国安卓应用,没有研究中国应用、微信小程序,也没有研究泡泡玛特(p. 3)。
No inter-rater reliability statistic is reported, so coding consistency cannot be judged (p. 4).论文没有报告评分者信度指标,因此无法判断编码的一致性(p. 4)。
Any app-favouring UI was coded as a DP, which may inflate the 95% figure (p. 4).凡对应用有利的界面都被记为暗黑模式,可能高估了 95% 这一数字(p. 4)。
The paper reports 589 respondents in the study and 584 in the conclusion, and education shares sum above 100% (pp. 8, 11).正文写 589 名受访者,结论写 584 名;学历比例加总也超过 100%(pp. 8, 11)。
Blind-box draws involve gambling-like randomness, which this taxonomy does not cover directly.盲盒抽取带有类似赌博的随机性,这套分类法并未直接涵盖。
Questions for the group · 小组讨论问题
Is a Pop Draw countdown a dark pattern if the stock limit is real?如果库存限制是真实的,Pop Draw 抽盒机的倒计时还算暗黑模式吗?
Should our audit code intent, or follow the paper's rule of coding any app-favouring UI?我们的审计应考虑设计意图,还是沿用论文「凡利于应用即记录」的规则?
Which dark pattern types would matter most for young Pop Mart fans and for regulators?对泡泡玛特年轻粉丝和监管者来说,哪类暗黑模式最要紧?
Written by the group from the full text. Numbers and page references were checked against the source. Licensed cases and reports are summarised, not reproduced.
由小组根据全文整理,数字与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。