基于归结的最大一般理论特化

RESOLUTION BASED MOST GENERAL SPECIALIZATION OF THEORY

提出一种基于归结的并有关于背景知识及示例的一致特化理论，该理论给出了最大一般特化假设的归结构造方法，可将其作为一种蕴涵意义下的一般理论特化框架．基于该理论，进一步提出ｋ 一般特化概念以解决特化的可计算性问题，并相应地给出特化算法．有关实验表明。

The acquisition or improvement of knowledge by learning is still one of essential issues in knowledge engineering. Inductive Logic Programming is such a learning method that makes use of background knowledge and learning examples in acquisition or improvement of knowledge. At present, the specialization of inductive hypotheses, one of main operations of the method, is limited to a few classes of inductive hypotheses and does not meet the principle of most generalization. Based on resolution and related to background knowledge and learning examples, this paper presents a theory of consistent specialization for the problem, which shows a way to construct the most general specialization of inductive hypotheses. This theory can be considered to be a general framework of theory specialization under logic implication. Furthermore, this paper sets forth, with the theory, a concept of k general specialization and corresponding algorithm to cope with the computability of the problem. The experiment shows that the theory specialization can be correctly and effectively computed with the theory and algorithm presented in this paper.