These can be derived from Markov chains (the probability of the next action given one). These can be determined from past usage as well.