In Classic Personalization, a "penalty" is applied to facet values that appear too frequently across all unique users. This is designed to filter out values that don't help differentiate between user preferences.
The penalty is based on the probability of a facet value appearing across all users. If a value appears for every user (ie. has a 100% probability), its score is reduced to 0. In this case, the value is considered "uninteresting" for personalization, since it is shared by everyone.
In short, if all users have interacted with the same facet value, that facet value will score 0 and so is dropped from individual user profiles.