A neural network model for hierarchical multilingual text categorization

Abstract

Enabling navigation via a hierarchy of conceptually related multilingual documents constitutes the fundamental support to global knowledge discovery. This requirement of organizing multilingual document by concepts makes the goal of supporting global knowledge discovery a concept-based multilingual text categorization task. In this paper, intelligent methods for enabling concept-based hierarchical multilingual text categorization using neural networks are proposed. First, a universal concept space, encapsulating the semantic knowledge of the relationship between all multilingual terms and concepts, which is required by concept-based multilingual text categorization, is generated using a self-organizing map. Second, a set of concept-based multilingual document categories, which acts as the hierarchical backbone of a browseable multilingual document directory, are generated using a hierarchical clustering algorithm. Third, a concept-based multilingual text classifier is developed using a 3-layer feed-forward neural network to facilitate the concept-based multilingual text categorization. © Springer-Verlag Berlin Heidelberg 2005.