(Online Demo) AdsDP: A Video Dataset for Recognizing and Examining Dark Patterns in iOS In-App Advertisements
Shang, Yuxuan
- Publisher
- Zenodo
- Language
- en
Abstract
Update on July 10, 2025: + We are uploading the whole AdsDP dataset to Kaggle (https://www.kaggle.com). This repository is an online demo that includes videos recorded by one researcher interacting with 81 English-language apps (which leads to 81 videos), and 9 Chinese-language apps (leads to 9 videos) along with the corresponding annotation files, in which you can directly view or download videos to get an overview of the dataset without having to download the compressed files. The complete dataset contains 718 videos, with a total duration of approximately 60 hours, and annotates 5,782 instances of ads-related dark patterns. Due to repository size limitations, we had to split the complete dataset into 2 parts (English apps and Chinese apps) for uploading. The links to the full AdsDP dataset are listed below: Online Demo: AdsDP: A Video Dataset for Recognizing and Examining Dark Patterns in iOS In-App Advertisements En-App: (En-App) AdsDP Cn-App: (Cn-App) AdsDP If you use this dataset, or the findings from the paper, please cite: @article{shang2025adsdp, title={AdsDP: A Video Dataset for Recognizing and Examining Dark Patterns in iOS In-App Advertisements}, author={Shang, Yuxuan and Wang, Guanxiao and Ren, Mengxia and Zhang, Haomin and Chen, Xingming and Xu, Haitao and Yue, Chuan and Hao, Shuai and Zhou, Bo and Ma, Wenrui and others}, journal={Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies}, volume={9}, number={3}, pages={1--31}, year={2025}, publisher={ACM New York, NY, USA} } ----------------- Update on May 1, 2025: + We added a manual for guiding researchers in operating the app and annotating ad dark patterns. + We added a record of a regular meeting, which helps to understand how we refined the codebook and manual, and clarified confusing instances to precisely define boundaries between ads dark patterns. ----------------- First Submission on Feb 1, 2025: This is a video dataset for studying deceptive or malicious UI designs related to advertisements, also known as "ads dark patterns". It captures the full interaction process of four researchers with 485 Chinese or English apps using iOS screen recording and annotates the occurrences and characteristics of identified ads dark patterns. More details about the dataset are provided in our paper, "AdsDP: A Video Dataset for Recognizing and Examining Dark Patterns in iOS In-App Advertisements." Each video corresponds to multiple rows in the annotation file. The first row contains the app’s basic information and dark patterns related to the ads' contextual design (namely, app's design), which are referred to as external dark patterns in the paper. From the second row onward, each row represents an ad and the dark patterns occurring within it, referred to as internal dark patterns. For more details, readers can refer to README. The complete dataset contains 718 videos, with a total duration of approximately 60 hours, and annotates 5,782 instances of ads-related dark patterns. Due to repository size limitations, we have currently uploaded videos recorded by one researcher interacting with 81 English-language apps (which leads to 81 videos), and 9 Chinese-language apps (leads to 9 videos) along with the corresponding annotation files. The complete dataset will be released soon.
Full text
Record of One Regular Meeting (Corresponding to §4.4 in the Paper) The following is an anonymous record of one regular meeting during the annotation phase, illustrating how majority voting and defining dark patterns through concrete instances were implemented and enforced. Time: 14:00, 17 November 2024 Question (1) from Researcher 3: An ad displayed the prompt “Swipe to claim a reward / to learn more.” After I swiped, the ad redirected to a landing page. Does this qualify as “Gesture-Induced Ad Redirection?” -Discussions: - Viewpoint 1 from Researcher 3: I believe this constitutes a dark pattern. The full-screen ad sets the entire screen to trigger a redirection upon swiping, making it very easy to activate while attempting to click the close button, despite I have taken extra care after noticing the prompt. - Viewpoint 2 from Researcher 2: I don’t consider this a dark pattern, as the ad has already provided a prominent warning. -Voting: Researcher 3 supported Viewpoint 1, while Researchers 1, 2, and 4 supported Viewpoint 2. -Decision: We cautiously decided not to consider it a dark pattern. Question (2) from Researcher 1: An ad displayed multiple close buttons simultaneously, and clicking any of them successfully closed the ad. Does this qualify as “Multiple Close Buttons?” -Discussions: - Viewpoint 1 from Researcher 4: This constitutes a dark pattern. As a user, I would hesitate because I don’t know in advance whether a button is functional, which indicates a flaw in the UI design. Additionally, I want to ask Researcher 1 how he determined that all the buttons were functional. - Researcher 1’s response: Ads with the same formats appeared multiple times, and each time I clicked a different close button, it successfully closed the ad. - Viewpoint 2 from Researcher 1: This is not a dark pattern. Although users may hesitate, clicking any of the buttons consistently produces the expected outcome of closing the ad. - Voting: Researcher 1, 2, 3 supported Viewpoint 2, Researcher 4 agreed with Viewpoint 2 after the discussion. -Decision: It is not a dark pattern. Question (3) from Researcher 1: Previously, we defined the dark pattern “Ad Reappearance” as the repeated appearance of ads in the same banner after being closed. However, during the actual annotation process, I found it difficult to determine whether two ads occupied the same banner, as it was unclear whether the banner was reloaded when the app interface changed. -Discussions: - Viewpoint 1 from Researcher 1: Perhaps we should define “ads in the same banner” as ads displayed in the same position and with the same size on the screen, and establish a minimum time interval between two ads below which they are considered the dark pattern “Ad Reappearance.” - Viewpoint 2 from Researcher 2: I suggest removing “Ad Reappearance.” Firstly, ads that refresh frequently overlap with “Long Ad/Many Ads.” Moreover, there is no public standard discussing an appropriate time interval for ads in the same banner. Furthermore, if ads within the banner cannot be closed or are frequently refreshed, it becomes difficult to determine which ones qualify as newly appeared ads when the app switches between different interfaces. -Voting: Researcher 2, 3, and 4 supported Viewpoint 2, Researcher 1 supported Viewpoint 1. -Decision: We removed the dark pattern “Ad Reappearance” from codebook. 1
2•Anon. Question (4) from Researcher 4: Some apps promote themselves to users, such as by recommending products they are selling. Does this type of design qualify as advertising? -Discussions: - Viewpoint 1 from Researcher 1: I have summarized several cases that should not be considered ads: (1) shopping apps promoting products they sell; (2) apps promoting their own features or making comparisons with similar apps. However, the following cases should be considered ads: (1) apps displaying products unrelated to themselves; (2) recommending other apps developed by the same developer. -Voting: All Researchers agreed with Viewpoint 1. -Decision: We will distinguish between regular content and ads based on these criteria.
