New Fitness Function for Ranking First-order Rule Based on Binding

In most problems of first-order rule learning, rule space is usually structured by thetas-subsumption operator. But in first-order rule space, thetas-subsumption is a quasi-ordering. If the number of coved training examples is used as the criterion for ranking candidates of hypothesis, there is a equivalent-class problem when searching along the quasi-ordering. Rules in an equivalent-class can't be distinguished according to their fitness function values. In another aspect, it makes the search to prefer longer rules, and would reduce the system efficiency and readability of learned rules. To solve these problems, in this paper, a new fitness function based on binding is presented. The contrast experiment has been done to show the effect of the new fitness function in guiding the search through first-order rule space