Oh, the Places You’ve Been!
Digging deeper into longitudinal tracking transparency.
Nathan Reitinger, Lilly Hackworth, Annika Hildebrandt, Hazel Murray, Bruce Wen, Michelle Mazurek, and Blase Ur
In 2012, Target ramped up its practice of tailored advertising by intently focusing on pregnant women, providing these women with a plethora of baby-themed advertisements. Unremarkable as this may seem today, the practice at the time was anything but — these women were targeted not based on user-provided demographic information or user-choice, opt-in advertising, but based on the purchase of 25 peculiar products.
'scent-free soap'
'calcium'
...
+ 'lotion'
--------------------
pregnantWhat may be more troubling than the reductionism this created was that Target — despite knowing not all customers would be “pleased” with this type of tailored advertising — did not change its behavior even after catching wind of consumer unease. Instead, the supermarket conglomerate sought to further insulate itself from reprimand:
With the pregnancy products … we learned that some women react badly . . . . [T]hen we started mixing in all these ads for things we knew pregnant women would never buy, so the baby ads looked random. <full story here>
But what would customers say if you asked them about this type of behavioral-based advertising?
More concretely, is there a point at which tracking has gone too far, a point when people start changing their behavior in order to avoid the side effects of tracking; are there some types of behavioral tracking that step over the proverbial “creepy” and end up inciting action in the form of tracker-blocking, ad-blocking, “private” browsing, and other privacy-enhancing technologies?
In this work, researchers at the University of Chicago and the University of Maryland set out to answer these questions in a follow-up study based on Oh, the Places You’ve Been! That study, published at the 2019 ACM SIGSAC Conference on Computer and Communications, found that users had a more accurate understanding of the tracking ecosystem when using a browser extension that made transparent hidden tracking practices. Here’s a screenshot of what the browsing extension looked like:
Although more than 90% of participants felt “surprised” in response to the breadth or depth of tracking when using the full extension, user attitudes regarding tracking did not differ substantially after using the tool. Paradoxically, it seemed that users found tracking creepy, but were OK with tracking occurring (see Figure 5 in the original paper).
In short, the visualization of tracking practices was interesting, but not galvanizing.
We aim to reexamine this finding: Would users have a more action-provoking response if the tracking visualizations were more invasive? To find out, we are building a new browser extension with what we hope will be more meaningful visualizations, in four categories: About You, Unique You, Sensitive You, and Advertisement History. Currently, we are feature testing a total of 22 visualization prototypes across these four categories to see which are the most meaningful to users.
About You
The About You section provides the user with information about how a tracker might view them. This includes demographic information, such as a predicted age and gender, personalized inferences pulled from Google AdSettings (e.g., inferred interests like breakfast foods or American football), and a heatmap of online activity suggesting the best times to advertise to this user.
Additionally, we provide our best guess as to where each inferred attribute came from. For instance, if Google said that a user was likely “not a parent” then we’d guess that this occurred because the user did not visit popular parenting websites, like themombeat.com or rookieparenting.com. Likewise, if we found that a user was likely in the 55–64 age category, then we’d guess that this user visited websites associated with that age, like the drudgereport.com (notably, our guesses will eventually be filled with real browsing history, but for the sake of these prototypes, we work from a fictional profile we’ve created).
Unique You
This category aims to highlight information about a user which has the least overlap with other users. We hypothesize that unique facts have more salience than common facts in clarifying the extent of tracking. Our Unique You visualizations highlight visits to unpopular or obscure websites, time-of-day oddities (i.e., how late you stay up online), and surge activities (i.e., heavy search history into specific topics like “finding a new job” or “relationship-heavy searching”).
Sensitive You
Certain types of web browsing habits are privacy-sensitive or perhaps even stigmatized, making this an area where users may feel like a line should be drawn. We test this hypothesis with visualizations aimed at potentially sensitive website visits (e.g., christianmingle.com) and interests (e.g., reproductive health or counseling services). We then pair those insights with the perspective trackers may have gleaned from these activities.
Ads
Advertising is often the end result of tracking. As such, we included a set of visualizations dedicated to uncovering the “why” behind the advertisements users see. Ads themselves sometimes provide explanations for why certain ads are served to certain individuals; however, these explanations are commonly ambiguous and may appear incomplete (e.g., you saw this ad because of “your similarity to groups of people the advertiser is trying to reach, according to your activity on this device” and the “websites you’ve visited”).
Instead, we aim to provide users with: (1) a set of visualizations synthesizing the advertisements they’ve seen; and (2) links between advertisements and a user’s browsing history. Here’s an example:
Although we are still in the midst of the preliminary prototype evaluation, the results so far are promising, with users generally feeling that the extension would help them better understand the implications of their online habits. Most users were “creeped out” by one or more of the visualizations, and many mentioned how this amount of detail was action-inspiring.
Additionally, so far, there appears to be a link between feeling a need to take action — i.e., the tracker has “gone too far” for this user — and the identification of a previously unknown type of tracking. For example, one user was surprised to learn that Google classifies users in a parental category (e.g., “Not a parent”) while another was surprised that tracking could be impacted based on the time of day (e.g., inferring and categorizing a user’s interest based on the time of day). These revelations gave rise to feelings of the need to change online behaviors.
Summary
Overall, we hope to illuminate the invasive ways users are tracked online with new visualizations and gauge how these visualizations affect user perceptions of those practices. After development of the finalized visualizations is complete, we will conduct a longitudinal study assessing user attitudes and perceptions of online tracking.
We gratefully acknowledge support from the Data Transparency Lab and Mozilla, as well as from a UMIACS contract under the partnership between the University of Maryland and DoD.
