Self-Organizing Maps for Classification of a Multi-Labeled Corpus

A Self-Organizing Map was used to classify the Reuters Corpus, by assigning a label to each of the documents that cluster to a specific node in the Self-Organizing Map. The predicted label is based on the most frequent label among the training documents attributed to that particular node. Experiments were carried out on different grid sizes (node numbers) to determine their influence on classification results. Informative visualizations of the resulting Self-Organizing Maps are demonstrated. We argue that the Self-Organizing Map is well suited to classify a document collection in which many documents simultaneously belong to several categories.