Survey of Large-scale Hierarchical Classification
Ankita A. Burungale, Dinesh A. Zende · 2014
Large-scale classification taxonomies have thousands of classes, deep hierarchies and skewed category distribution over documents. Hierarchical classification can speed up the classification process because problem is sub divided into smaller sub problems, and each of which can be efficiently and effectively managed. Most commonly used method for multiclass classification is one versus rest method. It is inflexible due to computational complexity. The top down method is usually accepted, but it has an error propagation problem. The metaclassification method solves error propagation problem. In this paper, several challenges for hierarchical document classification such as scalability, complexity, and misclassification are reviewed. The questions concerning about the learning and the classification processes are reviewed.