29.01.2014 Views

GWC 2008

GWC 2008

GWC 2008

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

SemanticNet: a WordNet-based Tool for the Navigation of Semantic… 29<br />

3.2 Evaluation of the Information<br />

Concerning the information added in the SemanticNet we have to distinguish between<br />

information gathered from WordNet and information extracted from Wikipedia<br />

documents. Properties of the information coming from WordNet are maintained in the<br />

structure used in the SemanticNet. Then we add new terms by means of a phase of<br />

classification of Wikipedia documents, in order to identify correctly information<br />

added in the map of concept.<br />

As we said, COMMONSENSE relations added in the SN are the links contained in<br />

the Wikipedia pages related to a particular sense of a term. The main question is to<br />

identify which categories are associated with that specific sense. We conducted our<br />

tests on sets of documents, extracted from a total set of 47639 documents and<br />

evaluating only 5 categories : Plants, Medicine, Animals, Geography and Chemistry.<br />

In the evaluation we only consider if the classified document belongs or not to the<br />

specified category. The Classifier assigns a number of possible categories to each<br />

document with a weight associated. We selected the best result for each document by<br />

means of a minimum level of weight. In this way all terms added to the SemanticNet<br />

are always related in the correct way with others.<br />

Fig. 2: Classified documents for each category<br />

In Fig.3 the measures of the Classifier about 5 categories are showed. Results are<br />

validated by hand verifying all documents identified by the Classifier for each<br />

category.<br />

Fig. 3: Measures of the classification

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!