As commenters have noted this has been done before. Likely, the reason why this can't be "fixed" is that it's a result of the curse of dimensionality [1] in the

As commenters have noted this has been done before. Likely, the reason why this can't be "fixed" is that it's a result of the curse of dimensionality [1] in the statistical sense: more dimensions make it exponentially easier to separate the space. This quote from the research suggest it as well, the error rate keep being divided by two while adding dimension (=interests) linearly: > the formula estimated that ’12, 18, 22, and 27 random interests make a user unique on FB with a probability of 50%, 80%, 90%, and 95%, respectively’. [1] https://en.m.wikipedia.org/wiki/Curse_of_dimensionality https://en.m.wikipedia.org/wiki/Curse_of_dimensionality