Abstract
Though a lot of work has been done to classify news headlines into generic categories like politics, sports, disasters, and so on, very little has been done to further classify a particular class of news into a deeper level of categories (especially in the case of disaster news). So, this paper presents a hierarchical classification of disaster news into appropriate disaster categories, based on their headlines. To demonstrate
effectiveness of such classification, a hierarchical classifier, trained using local classifier per parent node approach and in mandatory leaf-node prediction setting, is compared with its flat counterpart on disaster news classification, and no significant difference of performance in accuracy is observed. However, the use of a hierarchical classifier as a substitute is, then, justified by demonstrating its advantage in a non-mandatory leafnode prediction setting, where a significant difference of performance
in hP (hierarchical Precision) can be achieved by tuning the stopping criteria of the classifier.
Users
Please
log in to take part in the discussion (add own reviews or comments).