PowerCA: A Fast Iterative Implementation of Correspondence Analysis

The visual exploration of big data requires interactivity as well as the possibility to update an existing solution as new data becomes available in real time. Enhanced exploratory data visualization is provided by dimension reduction methods. Eigenvalue and singular value decompositions are the core of most of the dimension reduction techniques, such as principal component analysis (PCA) and multiple correspondence analysis (MCA). An efficient implementation of MCA, called PowerCA, is proposed that exploits enhanced computations of the sparse matrix transformations and fast iterative methods provided by intelligent initializations in case of repeated analyses. The aim is to extend the applicability of MCA to computational demanding application such as streaming text and web-log data visualization as well as bootstrap-based sensitivity analysis.