Construction d'un schéma d'implication entre variables issues d'une analyse des correspondances multiples. Développements. Application

[摘要] One of the limits of multiple correspondence analysis applied on large tables of qualitative items, is the difficulty of analysis and interpretation on the structure of relations among variables. In order to outpass the descriptive frontier, we propose a research methodology on involvement schemes which is based on conditional frequencies given by Burt's tables. The multiple correspondence analysis is used as a principal filter for variables from which (and as part of explanatory methods) oriented graphs are buit-up. They are accompagnied by some indicators (function, influence direction, driving force, besides one of the mass), which provide a greater legibility of the variables involved, and open the way to other advanced statistical analysis.

[学科分类] 数学（综合）

[关键词] terms clustering;knowledge acquisition;ontology;concept learning;corpus analysis;statistical data analysis;text-mining