Consideration on sensitivity for multiple correspondence analysis
Masaaki Ida · 2010
Abstract – Correspondence analysis is a popular research technique applicable to data and text mining, because the technique can analyze simple and multiple cross tables containing some measure of correspondence between the rows and columns and also has the strength of graphical display in two or three dimensions. The results of correspondence analysis provide information similar to factor analysis. Those abilities will deepen the global understanding on the characteristics of accumulated numerical and text information and have possibility leads to new knowledge discovery. We have so far focused on the text information of Japanese higher education institution such as syllabuses. We conducted research on collecting and analyzing the information, and on text mining to analyze and visualize the information for grasping the global characteristics of curricula of universities. Moreover, we have considered the sensitivity of ordinary simple correspondence analysis in case of data perturbation. However, more considerations on multiple correspondence analysis are required. This article presents mathematical considerations on sensitivity of multiple correspondence analysis.