Reading the article on which this is based (https://arxiv.org/abs/2110.06636 https://arxiv.org/abs/2110.06636), the researchers don't appear to specify how they

Reading the article on which this is based (https://arxiv.org/abs/2110.06636 https://arxiv.org/abs/2110.06636), the researchers don't appear to specify how they randomly select the 25 interests from the interests assigned to a user. This could have a considerable effect on their results, leading to a number somewhat higher than 22. The random selection could be uniform or it could be weighted by the global popularity of the interests. Presumably it's uniform. But if so, you could challenge the reasonableness of that assumption since it would be more likely to know a more common interest of a user that one in the tail of the distribution. For example, suppose I have three interests, two of which are very common and one of which is extremely niche. If you select two of these interests uniformly at random, chances are you'll get the niche one and be highly likely to identify me. OTOH if you randomly select weighted by the global popularity of the three interests, chances are that you'll get the two common interests and be very unlikely to identify me.