Direct clustering of a two-mode binary data-matrix

In this paper a discrete, categorical model is proposed for two-:­ mode data matrices with binary entries Xij E {0, 1}. The method operates directly upon a raw input (objects by attributes) data array by determining two-mode submatrices whose entries entirely have values x;j = 1. In addition, different submatrices are required to have a minimum set of objectattribute- pairs in common. Objects and attributes are classified simultaneously in a number ofjoint clusters. The method may be characterized as a non-hierarchical clustering procedure and in its model function to predict the data it is similar to Boolean factor analy is. Basically, the proposed method dubbed ' GRIDPAT' (for 'PATtern analysis of GRIDs') was developped for the structural representation of self concept data. Its relations to basic notions of Formal Concept Analysis and blockmodeling are discussed. An illustrative application of the approach is provided