Quantitatively Evaluating Formula-Variable Relevance by Forgetting

摘要

Forgetting is a feasible tool for weakening knowledge bases by focusing on the most important issues, and ignoring irrelevant, outdated, or even inconsistent information, in order to improve the efficiency of inference, as well as resolve conflicts in the knowledge base. Also, forgetting has connections with relevance between a variable and a formula. However, in the existing literature, the definition of relevance is "binary" -there are only the concepts of "relevant" and "irrelevant", and no means to evaluate the "degree" of relevance between variables and formulas. This paper presents a method to define the formula-variable relevance in a quantitative way, using the tool of variable forgetting, by evaluating the change of model set of a certain formula after forgetting a certain variable in it. We also discuss properties, examples and one possible application of the definition.