Extensive Experimental Evaluation of Self-Organizing Maps for Automatic Classification of a Multi-Class Multi-Label Corpus

摘要: This paper aims at bridging the gap between feature selection and space size by utilizing both square non-square self-organizing maps under different configuration scenarios for classifying a multi-class multi-label corpus, Reuters Mod Apte’ split. The of is based on heuristic process finding suitable map. Vector construction simple, yet effective procedure aiming transforming vectors from to uni-label. proposed solution improves classification efficiency not only in terms accuracy but also computational resources needed time training. Extensive experiments were conducted, using configurations regarding map vector sizes, training cycles, context words, assess their impact classifier’s performance. Furthermore, an intelligent algorithm label being proposed, show that neighboring nodes affect labels specific node. According our approach achieves 10% increase Macro-Average F1 scores, $30\times $ decrease dimensionality, approximately $34\times smaller when compared baseline scenario.