We already know that Facebook will create profiles for each user according to their activities, but the latest research from North Carolina State University found that this process does not seem to take into account the background of user activities.
For example, if you publish some content about green cheese that you don't like very much, the algorithm of Facebook's interest will notice that you share something about green cheese, but the algorithm will not record the context of your post: in fact, you don't like green cheese. Then, you may begin to receive targeted ads for green cheese.
To learn more about how Facebook generates interest profiles for users, researchers conducted two experiments.
In the first experiment, researchers created 14 new user accounts on Facebook. The researchers controlled the demographics and behavior of each account and tracked the interest lists generated by Facebook for each account.
This can check your Facebook account's advertising preferences. https://www.facebook.com/adpreferences/
The first experiment enables us to see which activities are related to Facebook's interest inference. The key discovery here is that Facebook has adopted a radical interest inference method.
Even with simple operations like scrolling the page, Facebook thinks users are interested in the topic. For the 14 accounts we created for this study, we found that 33.22% of the inferential interests were inaccurate or irrelevant.
Then the researchers conducted a second experiment to determine whether these findings apply to a larger and more diverse user group.
In the second experiment, the researchers recruited 146 participants from around the world. Participants downloaded a browser extension that allows researchers to collect data about each person's interests from their Facebook accounts.
This study found that 29.3% of the interests listed by Facebook for research participants were not interested.
The original paper in English is attached here: https://anupamdas.org/paper/CSCW2022.pdf