Parallel Web Text Clustering with a Modular Self-Organizing Map System

In this study, a multistage modular self-organizing map (SOM) model is proposed for parallel web text clustering. In the first stage, the large textual datasets are divided into some small disjoint datasets (i.e., task decomposition). In the second stage, each small data set is input into different unitary SOM models for word clustering map (i.e., modularization learning). In this stage, different SOM models are implemented in a parallel way to gain greater computational efficiency and scalability. In the third stage, based upon the outputs of each SOM module in the previous stage, another SOM model is used to integrate different word clustering results to formulate a text category map (i.e., module fusion). In the proposed model, word clustering map is embedded into text category map and thus a hierarchically modular SOM model is formulated. For illustration and verification purpose, a practical text clustering experiment is performed.