| 1: procedure FeatureSelection |
| 2: BOW ⟵ Bag of Word representation of the corpus |
| 3: L ⟵ Set of Leaders from Algorithm 1 |
| 4: G ⟵ The graph model from Algorithm 1 |
| 5: begin: |
| 6: W={} |
| 7: foreach Word w in BOWdo |
| 8: if w thenord ∈ HEF such as link text, headings, meta informations, image descriptions etc... |
| 9: W[w]=BOW[w]1.5 |
| 10: else |
| 11: W=[w]=BOW[w] |
| 12: DocClass={} - Represent the Document-Class vector, all documents are initialized 1 to all classes. |
| 13: foreach leader l in Ldo |
| 14: class,prob=get class and probability of l from the classifer |
| 15: update DocClass, value=l, class and prob, weight =1.5 |
| 16: foreach neighbor n of l in Gdo |
| 17: update DocClass, value=n, class and prob, weight =1.5 |
| 18: foreach document d in the corpusdo |
| 19: class,prob=get class and probability of d from the classifer |
| 20: update DocClass, value=l, class and prob, weight =1 |
| 21: Assign the class which is having the highest probweight value to d. |
