The effect of applying any of these three filters on actual data can be found in the evaluation section (section 9). A small sample from an ontology built using the n-gram learner after refinement is shown in figure 4. Figure 4 Part of the refined N-gram Ontology represented in English The basic idea of this algorithm is to extend the learnt ontology by locating concepts that appear at the same level in a document as a previously learnt concept. Having located any one of these, it is added to the ontology as a sibling to the previously learnt concept which appears with it at the same level in the document. The refinement algorithm also calls the previously developed refinement algorithm (section 7) in order remove concepts that do not frequently occur in headings as previously described. Figure 7 Part of Refined HTML Ontology represented in English A portion of the ontology learnt using the method based on HTML structure after refinement is shown in figure 7.
