I usually bundle feedback in two groups - first one is my target demographic, the second one is professional. First one is pretty straightforward to parse, the second one is a kind of a double-edged sword.

Whatever your methodology is, at some point you can't parse more feedback, and it's then noise, whatever its contribution is.

I see it this way: The higher frequency of exposure, the larger the sampling pool, the more noise AND the more opportunity to spot a recurring signal in that noise. If you keep your sampling pool low, how will you separate noise from signal?