Un metodo di intelligenza artificiale per l’estrazione di informazione da dati statistici

Abstract

The practice of knowledge discovery from massive statistical data by data mining is not outright yet. That mainly because the techniques that data mining itself objectively utilises, often are artful indeed; in addition , in general such techniques do not encompass semiotic aspects of the analyzed data, which certainly constitutes a limitation. The paper presents a tentative to increase data mining effectiveness in elaborating such data statistical dimensions on the basis of the data themselves semantic characteristics. That essentially consists in extracting conceptual information in function of the data semantic relevance also. The elaboration in topic is realized by analysing statistical data through the particular system HERMES. The gradient of typicality of such data’s components, with respect to the extraction aimed purposes, assumes a determinant role in the development proposed by the paper. To exemplify the fundamental ideas which HERMES is based on, a practical quite simple application is illustrated in the paper itself, through a paradigmatic example. Such application regards how to investigate on the causes of the different success degrees of television programs, on the basis of the programs composition as well as on the shares the programs themselves obtained; shares derived from analysing view figures regarding a large statistical survey recorded by devices like AUDITEL.