When scoring function is the Maximum Likelihood, the model would make the data as probable as possible by choosing the graph structure that would produce the highest score for the estimate of the parameter.