An unsupervised semantic similarity based method for word sense disambiguation

Abstract

In this thesis, a semantic similarity based unsupervised method for word sense disambiguation is presented. The method tries to disambiguate a target word by calculating a similarity score between the words surrounding the target word and the words existing in the sense definition of the target word. The built-in semantic hierarchy and synset relations of WordNet, a machine readable thesauri, are used in similarity score calculations. The method is evaluated using SemCor data and the results are compared against other methods based on semantic similarity and unsupervised methods. Results show us that increasing the number of inputs by including the words in a word’s sense into disambiguation process, improves precision rate of disambiguation process.