A fast dendrogram refinement approach for unsupervised expansion of hierarchies

Hierarchies are effective data models for organizing textual collections, particularly for automatic document classification into categories and subcategories. However, the majority of existing methods on hierarchical classification require human-labeled document set. Moreover, humans have good insight to manage the categories of higher levels of the hierarchy, i.e., more general categories, while the management of more specific categories is a difficult and expensive task since it requires expert knowledge to identify appropriate categories and their respective documents. Thus, in this paper we introduce an approach to automatically expand new, and more specific categories from a reduced initial hierarchy, which contains only general categories. Our approach is based on text clustering methods, particularly performing refinements on dendrograms obtained by hierarchical clustering algorithms. The results of the experimental evaluation show that the proposed approach achieves better performance in the expansion of hierarchies, compared with a traditional technique. Moreover, our approach is computationally faster, allowing the identification of new categories in large text collections.