Analyzing the Impact and Accuracy of Facebook Activity on Facebook's Ad-Interest Inference Process

Abstract

Social media platforms like Facebook have become increasingly popular for serving targeted ads to their users. This has led to increased privacy concerns due to the lack of transparency regarding how ads are matched against each user profile. Facebook infers user interests through their activities and targets ads based on those interests. Although Facebook provides explanations for why a particular interest is inferred about a user, there is still a gap in understanding what activities lead to interest inferences and the extent to which the sentiment or context of activities is considered in inferring interests. To obtain insights into how Facebook generates interests from a user's Facebook activities, we performed controlled experiments by creating new accounts and systematically executing numerous planned activities. This enabled us to make causal inferences about activities that lead to generating specific interests, many of which were not representative of actual user preferences. We also evaluated which activities resulted in interests and found that very naive activities, such as only viewing/scrolling through a page, lead to an interest inference. We found 33.22% of the inferred interests were inaccurate or irrelevant. We further evaluated the interest inference explanations provided by Facebook and found that these explanations were too generalized and, at times, misleading. To understand if our findings hold for a large and diverse sample, we conducted a user study where we recruited 146 participants (through Amazon Mechanical Turk) from different regions of the world to evaluate the accuracy of interests inferred by Facebook. We developed a browser extension to extract data from their own Facebook accounts and ask questions based on such data. Our participants reported a similar range (29%) of inaccuracy as observed in our controlled experiments. We also found that most of our participants were unaware of the availability of Facebook's ad preference manager, interest inference process, and even interest explanations.

How Does Connecting Online Activities to Advertising Inferences Impact Privacy Perceptions?

“…In particular, we ask participants to not only observe the labels applied to them by advertisers but also to imagine generating said labels. The second major area focuses on measurements within the advertising ecosystem [2,8,31,32,40], such as its accuracy and the frequency in which the inferences are updated or individualized. Our work builds on this line of research as well by observing directly how users expect labels to be applied compared to how they are actually applied as a proxy measurement of the accuracy of the labels.…”

“…To understand how visualizing inference-level information about online tracking impacts people's knowledge, perceptions, and attitudes, Weinshel et al [39] built a browser extension to conduct a longitudinal field study, finding that after participants viewed visualized examples, they had a more accurate perception of tracking and were more likely to take action. Sabir et al [32] explored how Facebook generates ad interests and found that even minor activity, such as scrolling through a page, can lead to an inference. But interests can often be inaccurate with explanations that are too generalized and/or misleading.…”

“…• collection-beneficial (319) personalized-ads (106), better-recommendations (85), personalization (57), search (46), improved-experience (34), youtube (22), revisit-activities (14), maps (10), autofill-forms (3), search-history (2), news (2), storing-information (2), free-products-services (2), store-passwords (2), google-rewards (1), sometimes (1), saved-logins (1), websites (1) • privacy-concerns ( 218) amount-of-information (62), selling-data (33), information-collection (27), tracking (24), sensitive-information (19), third-parties (18), monitoring ( 14), do-not-want-others-to-see (12), invasion-of-privacy (11), information-used-against-me (9), amount-of-interests ( 6), amount-of-knowledge (6), location (4), search-history (3), scary (3), targeted-advertising (3), creepy (3), ads-too-specific (2), specific-information (2), societal-consequences (1), stalking (1), whatinformation-collected (1), interests-too-accurate (1), behavior-modification (1), ads-everywhere (1), uncomfortable-sharing (1), how-long-data-stored (1), why-information-collected (1) • unconcerned (125) non-sensitive (31), low-risk ( 21), information-is-generic (11), interests (7), nothing-bad-has-happened ( 5), collection-meets-expectations (4), data-collection (4), comfortable-sharing (4), because-interests-not-accurate (3), interests-not-sensitive (3), has-read-privacy-policy (2), do-not-care-who-knows-interests (1), information-is-secured-by-google (1), information-collected-is-not-important (1) • no-benefit (116) no-use-ads (32), personalized-ads-not-helpful (6), interests-not-helpful (5), information-collected-not-helpful (3), no-use-recommendations (2), experiencenot-enhanced (2), causes-problems (1), recommendations-not-helpful (1), not-paid-for-data-collection (1), personalized-ads-invasive (1) • did-not-change-perspective (109) • some-interests-incorrect (54) one-off-activities (15), gave-examples (13), outdated (3), does-not-like-wrong-interests (2), is-okay (1), conflicting-interests (1), very-inaccurate (1), wants-todelete (1) • unknowns (38) how-information-used (17), who-has-access (10), what-information-collected (8), how-much-information (4), why-information-collected (2), how-longinformation-stored (1), where-information-goes (1), risk (1) • interests-not-accurate (32) only-one-search (3), wro...…”

Do you really know? Incidental exposure to pro- and anti-consensus science information on social media, knowledge overconfidence, and science skepticism

“…Users can partly customize their feeds by subscribing to the channels they like which increases the likelihood to be exposed to content from that channel in the future and social contacts may share contents from channels someone does not follow themselves or create novel content (Thorson & Wells, 2016). Because social media companies track users' (intentional) behaviors, algorithmic inferences about user interests are used to display content in the future (DeVito, 2017;Sabir et al, 2022). To illustrate, Thorson et al (2021) found that Facebook users who had been algorithmically classified as interested in news and politics were also more likely to be exposed to these contents.…”