I did a bit of messing around with this around Christmas. Nothing that changed the acceptance algorithm, just something to vary the responses according to the girl's mental state. There are a lot of factors that probably should be taken into account, and the amount of text grows quickly, the more factors you consider.
Also, and this was a surprise, the order in which you check attributes seems to affect the tone of the response. That is, if I write a decision table that checks PC Hate first, I seem to end up writing a set of responses that are primarily about how she hates the player. If I consider libido first, the responses take a different tack entirely. Of course, it doesn't have to be that way; it was just the way I found things working out. Still, I thought we could use this to advantage.
The thing to do, I think, would be to consider at most four stats, and then to randomly select the stats checked, but weighted by how strong her feelings were. So if a girl had 100 hate, you'd tend to see a lot of angry responses, but it would be possible to catch her in a good mood from time to time, for instance.
The other thing I might look at, (and this will not happen for some time yet) would be to try and work out a format for a response template. So you could design reponse matrices based on various stats, and with custom text and probably accpetance tests. It would give people a lot of control over how girls reacted, and using templates that could be shared between girls would cut down on the work involved.
But like I say, not for some time, and in any case I want to think about it a bit more.