Word sense disambiguation in evolutionary manner

The task of assigning proper meaning to an ambiguous word in a particular context is termed word sense disambiguation (WSD). We propose a genetic algorithm, improved by local search techniques, to maximise the overall semantic similarity or relatedness of a given text. Local search is used because of the inefficiency of population-based algorithms (e.g. genetic algorithm) in exploiting the search space. Firstly, the proposed method assigns all potential senses for each word using a WordNet sense inventory. Then, the improved genetic algorithm is applied to determine a coherent set of senses that carries maximum similarity or relatedness score based on information content and gloss overlap methods, namely extended Lesk algorithm and Jiang and Conrath (jcn). The obtained results outperformed other unsupervised methods, which are related to the proposed method, when tested on the same benchmark dataset. It can be concluded that the proposed method is an effective solution for unsupervised WSD.