In that case, I think defining "attraction" on scale of 1-10 is more useful than binary classification.
maybe you should think about how you would use the data more carefully.
You are correct. Binary classification is probably the best way to prove that hypothesis.
Shame on me for not recognizing how silly a hypothesis it was sooner.
If you believe that hypothesis to be silly, then put forth your own hypothesis. Be sure to choose a hypothesis that allows you to use the data you want to use (which is a backasswards way of doing it, but whatever).
I don't need to put forth a new hypothesis to recognize one is silly.
However, in this case:
the hypothesis, if we are being forced to have one, is that 90+% of male humans are 99+% homogeneous in the sex of other humans they are attracted to.
I think that if you don't attempt to prove homogeneity and instead grade all possible attractions on a sliding scale (that as far as I can tell, everyone agrees is present), you'd have far more interesting/useful results (even if the attractions are all from a single gender)
You could very well end up with similar results (90% of male humans are 99% homogeneous with regard to sexual preference), but you open up a lot of interesting possibilities that would unavailable in a binary classification. I still think grading on a 1-10 (or 1-7 scale) is an oversimplification, but it's better than "would you do her or not".