Automatically learning document taxonomies for hierarchical classification
Kunal Punera, Suju Rajan, Joydeep Ghosh · 2005
While several hierarc hic al c assific ation methods have been applied to web c ntent, suc htec hniques invariably rely on a pre-defined taxonomy of doc uments. We propose a new tec hnique that extrac ts a suitable hierarc hic al struc ture automatic ally from ac orpus of labeled doc uments. We show that our tec hnique groups similarc lasses c oser together in the tree and disc overs relationships among doc uments that are not en c ded in the c ass labels. The learned taxonomy is then used along with binary SVMs for multi-c lassc lassific ation. We demonstrate the e#c ac of our approac h by testing it on the 20-Newsgroup dataset.