privacy-tech-lab / privacy-pioneer

Privacy browser extension for analyzing web traffic of visited websites
https://www.privacytechlab.org/
Other
26 stars 1 forks source link

Why are GPS coordinates for my visit of nba.com labeled as coarse location instead of fine location? #509

Closed SebastianZimmeck closed 1 year ago

SebastianZimmeck commented 1 year ago

As the title says. My GPS coordinates are correct, but it would seem more appropriate to me to label them as fine location. Are we always labeling GPS coordinates as coarse? Is this one-off for nba.com? Also, if we change any labeling category here, do we potentially confuse the original machine learning categorization?

Screenshot 2023-06-20 at 10 06 57 AM
SebastianZimmeck commented 1 year ago

Also, a bit along the same lines, it is not clear to me why the below label has both coarse and precise location. It looks pretty much the same to me as the label for nba.com.

Screenshot 2023-06-20 at 10 32 13 AM
SebastianZimmeck commented 1 year ago

Maybe, just a labeling error (that we would expect once in a while according to our performance numbers) of nba.com?

danielgoldelman commented 1 year ago

We use the Browser Location API, which is built in to Firefox, to obtain a user's location. We save these on the user's extension and compare against those values in PP. Fine location means that the number calculated by the individual website is within +-0.1 degrees from the Browser Location API value, and Coarse Location means that it is within +-1.0.

Therefore, I assume that the value calculated by nba.com (maybe using your IP instead of the Browser Location API) was within the coarse location bound, but not the fine location bound.

The second example could thus be within both bounds, and therefore generating both labels.

I would recommend a change to the main README to reflect these distinctions, and not describe this as an error.

SebastianZimmeck commented 1 year ago

I would recommend a change to the main README to reflect these distinctions, and not describe this as an error.

Fine location means that the number calculated by the individual website is within +-0.1 degrees from the Browser Location API value, and Coarse Location means that it is within +-1.0

Yes, let's include this in the readme on how the location API works. This should be documented. Especially, intuitively any lat/lon data may be perceived as fine location. It is true that the attempt may fail, but the intent is there just evidenced based on the format.

This should also be described in the paper in the implementation section. I included a todo item there.