Русский
!

Conference publications

Abstracts

XVII conference

An algorithm of the detection and the account of latent relationships in regularities of the knowledge base

Karp V.P., Sarookhanov B.A., Safarova E.A.1

The Moscow State Institute of Radioengineering Electronics and Automation (technical university), Cybernetics department, Information Systems subdepartment, Russia 119454 Moscow, Vernadsky avenue, 78, phone: (903)2166547, e-mail: karpvica@mail.ru

11 Medical Centre “Meda-21”, Moscow

1 pp. (accepted)

In the process of solving classification tasks referring to complex-organized objects there is a necessity of the additional analysis of the constructed diagnostic rules, i.e. knowledge base (KB). Complex-organized objects are characterized by presence of a large number of qualitative features and presence of the missing data generated, frequently, by impossibility of direct measurement of the interesting indicator. Due to the latter, researchers have to include indirect features with probable connectivity, which may misrepresent the result of computer diagnostics considerably. These features are required to be considered while forming KB beyond the dependence on a method of classification rules construction.

The developed algorithm of “the account of the latent connections” is oriented towards the analysis diagnostic KB that is constructed by a search conjunction method (SCM). KB is presented in the form of a set (ordered on decrease) of diagnostically significant conjunctions of symptoms, typical for each class [1]. Each conjunction consists of two interconnected fragments: «numbers of symptoms» forming conjunction, and it’s «semantic meaning».

The idea of algorithm consists in the additional analysis of the KB under two circumstances. The first one concerns SCM ability to suppose several (no more than three) conjunctions with the same "leading" symptom. In this case the KB fragment consisting of numbers of symptoms is used for the analysis and the so-called K-restriction conditions is applied, according to which of a set of conjunctions with the same leading symptom for each concrete object the most diagnostically significant one is considered. The second situation is generated by availability of latent (indirect/unevident) substantial dependences available in the KB [2]. For their detection the algorithm counts up factors of pair correlations between all pair conjunctions available in the KB. The conjunctions which formed pairs with "high" correlations factors are exposed to the additional semantic analysis of maintenance and, if required, to the KB semantic transformation. Such approach allows to consider the given algorithm as one of the ways of the knowledge base quality improvement.

References

1. Karp V. P. Intelligent data analysis in a problem of decision rules construction (on an example of a medical diagnostics problem) // Artificial Intelligence News. – 2006. – №2. – p.57-75.

2. Karp V. P., Sarookhanov B. A. Forming of problem-oriented database in tha passive experiment conditions // Set of works from scientific and technical conference. “Information technology and computer facilities”. MSIREA. – M., – 2007. – p. 53-57.



© 2004 Designed by Lyceum of Informational Technologies №1533