Evaluation of semantic similarity in WSD: an analysis to incorporate it into the association of terms

• Erick Nilsen Pereira de Souza UFBA
• Daniela Barreiro Claro UFNA

Resumo

Semantic association requires an analysis on similarity in order to provide better combinations. Several semantic similarity measures have been proposed for the Word Sense Disambiguation (WSD) problem. Therefore, this work aims to evaluate the performance of five semantic similarity measures into a thesaurus by WSD approach so as to incorporate semantic similarities into association problems. Experiments were conducted using two main algorithms of WSD domain. The results showed that the application of semantic similarity improves precision and recall on disambiguation domain. In order to complement the evaluation, we proposed a hybrid approach and we evaluated it over other measures presented. According to our results, it was possible to adapt a similarity measure for the problem of finding semantic associations in texts.