Automatically Adjusting Content Taxonomies for Hierarchical Classification

Hierarchical models have been shown to be effective in content classification. However, the performance of the model heavily depends on the given hierarchical taxonomy. We empirically show that different taxonomies can result in significant differences in hierarchical classification performance. Motivated by some real application problems, we aim to modify a content taxonomy automatically for different applications. In this work, we formulate the problem, discuss why it is feasible to achieve better performance in terms of classification performance via adjusting a given hierarchy, and present one effective solution to find better hierarchies compared with that of the given original hierarchy. Preliminary experiments on some real world data sets are reported and discussed.