Using features extracted from Wikipedia for the task of Word Sense Disambiguation

Abstract

In this paper, a method using features extracted from Wikipedia for the task of Word Sense Disambiguation (WSD) is presented and evaluated. A term-concepts table constructed from Wikipedia and the redirect links is described. With its help, the Wikipedia internal links along with the categories structure are used to compute the relatedness between any two concepts through a two-level process: a term-concepts expansion followed by a links-based expansion. The result is a ranked list of concepts which are most related to the ambiguous term given the context it exists in. For the evaluation experiment, the benchmark is constructed from a segment of the internal links of Wikipedia. The evaluation results obtained suggest that introducing links analysis and the categories structure to the built term-concepts table provide improvement to the accuracy of the method in the WSD task.