Enhancing Top-Down Classification Method with Metaclassification for Large-scale Dataset

Ankita Burungale, Dinesh Zende

Abstract: Large-scale hierarchical classification has thousands of classes. The most commonly used method for multiclassification is one-versus rest, which is inappropriate due to computational complexity. So, Top down Method is used instead, but it is not perfect because of an error-propagation problem. The error-propagation means the document are wrongly rejected at higher level, it cannot passed down. Metaclassification Method solves error-propagation problem but it has higher complexity. To overcome this problem, enhancing top method is proposed which combines Top down and Metaclassification methods. It uses score of all base classifiers along root to leaf and checks whether predicted label is correct label or not.

Keywords: Hierarchical text categorization, Metaclassification, Top down method, Meta level learning, Large-scale data

How to Cite?: Ankita Burungale, Dinesh Zende, "Enhancing Top-Down Classification Method with Metaclassification for Large-scale Dataset", Volume 4 Issue 6, June 2015, International Journal of Science and Research (IJSR), Pages: 712-716, https://www.ijsr.net/getabstract.php?paperid=SUB155204, DOI: https://dx.doi.org/10.21275/SUB155204

Download Citation: APA | MLA | BibTeX | EndNote | RefMan