Redirigiendo al acceso original de articulo en 17 segundos...
ARTÍCULO
TITULO

GENERATION OF TEST BASES OF RULES FOR THE ANALYSIS OF PRODUCTIVITY OF LOGICAL INFERENCE ENGINE

Svitlana Shapovalova    

Resumen

The subject of research in the article are test tasks to determine the performance of logical inference engines based on rules. The purpose of the work is to create a method of forming a database of rules and a set of data for analyzing the performance of logical inference mechanisms according to the given characteristics of rule activation and the complexity of finding a way to the target conclusion. The article solves the following tasks: determining the requirements for the knowledge base to be formed; creation of a knowledge base model; creating a way to form rules; identifying ways to increase the number of rules; providing testing of logical inference mechanisms for the proposed test problem. The following methods are used: methods of comparison with the sample, graph theory, logical programming. The following results were obtained: the method provides opportunities: creation of conditions of rules that complicate the data flow network of the Rete-algorithm as much as possible; formation of test bases of rules for derivation both on logic of the first order, and on offer logic; simply increase the number of knowledge base rules while maintaining the output logic. The formation of the knowledge base is based on a graph that represents the meta-rules of mixing paints to obtain a new colour. The vertices of the graph are colour classes. Each metarule is either an edge leading to the OR vertex or a set of edges in the case of the AND vertex. Each meta-rule specifies a scheme for creating several rules, because its structural components are classes of paints. The given structure of the graph significantly complicates the logical inference, because to prove the truth of the conclusion on AND-vertices it is necessary to have the conclusions obtained in the previous steps of different search directions.  Examples of rule formation are given. Target vertices are defined, which determine the simplest and most complex cases of logical inference. Conclusions: it was proposed a semantic model of the knowledge base in the form of AND/OR-graph, which allows you to test the effectiveness of the implementation of conflict resolution strategies, as well as heuristic algorithms; a method of creating tests for inference mechanisms, which allows you to generate a database of rules and a set of data of certain sizes, as well as to model the complexity of finding the target output and activating the rules. Ways to increase the number of rules of the knowledge base to complicate the problem of logical inference have been presented; formulation of tests to determine the performance of logic output mechanisms for the proposed test problem has been done.

PÁGINAS
pp. 77 - 85
REVISTAS SIMILARES

 Artículos similares