Which ML approach to take? Predicting real-valued (-1, 1) metric based on frequency of certain phrases.

I'm fairly new to machine learning, and was wondering if any one could offer advice. I have several hundred speakers, each having a vector of ~250 relative phrase frequencies (#of times given phrase is said/# of times any phrase in set is said). I want to predict a real-valued measure of ideology, with values ranging from -1 to 1. I was planning on using Support Vector Regression, but I was wondering if using something like a Multilayer perceptron might be better.