Self-Organizing-Map-Based Metamodeling for Massive Text Data Exploration

In this study, we describe the use of the self-organizing map (SOM) as a metamodeling technique to design a parallel text data exploration system. Firstly, the large textual collections are divided into various small data subsets. Based on the different subsets, different unitary SOM models, i.e., base models, are then trained for word clustering map. In this phase, different SOM models are implemented in parallel to gain greater computational efficiency. Finally, a SOM-based metamodel can be produced to formulate a text category map through learning from all base models. For illustration the proposed metamodel is applied to a massive text data collection.