I would like to implement a simple Bayesian classification system to do rudimentary sentiment analysis on short messages. Practical suggestions for implementing in Ruby would be
I've just released the nbayes gem which is a full-featured, robust, and high performance Naive Bayes classifier in Ruby. It makes tasks like sentiment analysis simple. Please read the tutorial for more info.
Ilya Grigorik has a nice answer to this problem over on this blog post on Bayesian Classifiers
Additionally, you may wish to take a look at the ai4r rubygem for some alternates to Bayesian Classifiers.
ID3 is a good choice because it gives a decision tree that is "understandable" to even someone without any real understanding of machine learning techniques.