Relationship between the Modularity criterion and the Relational Analysis

This paper studies the extension of the Modularity measure for categorical data clustering. It first shows the relational data presentation and establishes the relationship between the extended Modularity and the Relational Analysis criterion. Two extensions are presented in this work: the early integration and the intermediate integration approaches. The proposed Modularity measure introduces an automatic weighting scheme which takes in consideration the profile of each data object. An iterative algorithm is then presented to search for the partitions maximizing this criterion. This algorithm deals linearly with large data sets and allows natural clusters identification, i.e. doesn't require fixing the number of clusters and the size of each cluster. For the early integration approach, several experiments are conducted in order to show the effectiveness of the proposed approach.