AdEcologist: A Browser-based Tool for Crowdsourcing Data Collection of Ads and Ad Auctions on the Web
The online advertising ecosystem is extremely opaque, especially when it comes to targeted advertising. From the outside, we have very little data on how ads are targeted, such as who is being targeted by what, and how prevalent different forms of targeting (e.g. contextual, behavioral, retargeting) are. This lack of transparency makes it difficult to have informed discussion about tradeoffs between targeted advertising and user privacy.
Studying targeting without data from the ad tech companies is difficult. Because of the sheer volume of ads on the internet, and the multitude of parameters advertisers can use to target people, one must collect a large amount of data from multiple browsing profiles to make inferences about how a particular ad or category of ad is targeted. And to capture truly ecologically valid data, the ads ideally are collected from real users' browsers, rather than web crawlers.
AdEcologist is a browser extension for crowdsourcing data collection of ad content on the web. It can scrape screenshots of ads on the websites that the user visits, as well as information on header bidding ad auctions (for sites that have them). Researchers can distribute a variant of AdEcologist to study participants or volunteers to collect a dataset of ads from a diversity of vantage points.
This (base) version of AdEcologist is not just a scraping tool, but also includes user interfaces to guide study participants or volunteers through the setup and usage of the extension. Developed for a previous user study [1], the extension includes flows for:
- Registering the extension with the researchers' backend
- Brief introduction and tutorial
- To-do list of websites to visit
- A survey asking participants to rate the relevance of the ads shown to them
- An interface to allow participants to review and withhold any ad screenshots that they may consider too privacy-sensitive to share with researchers
The dataset that AdEcologist generates can be used to study ad targeting using a "topic-based" or "content-based" approach. In our previous work [1], we used OCR, topic modeling, and manual qualitative analysis to sort ads into categories, and then analyzed how different ad categories were distributed across demographic characteristics and websites.
AdEcologist also contains functionality for collecting data on ad auctions conducted in header bidding scripts. This data provides information on the value of bids offered to fill ad slots by different ad networks, and the value of the winning bid. This information can be used to analyze how much advertisers value a particular user's impression, or a particular website. It can also suggest when an ad is a retargeted ad - research findings show that retargeted ads fetch significantly higher bid values [2,3].
Disclamer: This tool is research code, and may not work out of the box for all use cases. However, we hope that it provides a starting point for other researchers hoping to collect ad data with the help of real users. We welcome contributions, and encourage modifications for your own studies. If you find this codebase at all helpful for your efforts, consider citing our paper [1].
This repository contains two components - the Chrome extension and a backend. See the extension README and server README for more details on setting up the infrastructure.
A Chrome/Chromium extension for crowdsourced data collection of ad content from the web. Contains user interfaces suitable for a user study, but can also configured to operate more like a crawler.
A simple node.js backend for receiving the data from the extension. The data is more or less directly stored in a PostgreSQL database. It also contains basic authentication/registration features to restrict usage to registered study participants.
AdEcologist was developed by Eric Zeng and Rachel McAmis. It was used to collect data in our IMC 2022 paper measuring targeted advertising and ad auctions, and how they vary across demographic characteristics and websites [1].
This work is supported by the National Science Foundation under awards CNS-1565252, CNS-1651230, and CNS-2041894, the John S. and James L. Knight Foundation, the Consumer Reports Digital Lab Fellowship, the UW Center for an Informed Public, and the UW Tech Policy Lab, which receives support from: the William and Flora Hewlett Foundation, the John D. and Catherine T. MacArthur Foundation, Microsoft, the Pierre and Pamela Omidyar Fund at the Silicon Valley Community Foundation.
[1] Eric Zeng, Rachel McAmis, Tadayoshi Kohno, Franziska Roesner. What Factors Affect Targeting and Bids in Online Advertising? A Field Measurement Study. ACM Internet Measurement Conference (IMC), 2022.
[2] Lukasz Olejnik, Tran Minh-Dung, Claude Castelluccia. Selling Off Privacy at Auction. Network and Distributed System Security (NDSS) Symposium, 2014.
[3] Michalis Pachilakis, Panagiotis Papadopoulos, Evangelos P. Markatos, Nicolas Kourtellis. No More Chasing Waterfalls: A Measurement Study of the Header Bidding Ad-Ecosystem. ACM Internet Measurement Conference (IMC), 2019.