An Empirical Analysis of Word Sense Disambiguation through Machine Learning Approaches

Abstract

The procedure to identify the appropriate meaning for the particular word in an ambiguous statement is considered as Word Sense Disambiguation. It is a complicated problem since it necessitates the utilization of information from a variety of sources. Since the start of machine learning, a significant amount of time and effort has been devoted to overcoming this challenge, and the work is currently ongoing. In WSD, a variety of methodologies were employed and executed on a variety of corpora representing practically all languages. WSD algorithms are grouped into three groups in this paper: Supervised algorithms, unsupervised algorithms and knowledge-based algorithms. Every subcategory will be examined thoroughly, with details elaborated for nearly all of the algorithms within each area. As a result, work samples for every technique were selected based on the language being used, the corpora being used, and other considerations. Each method's advantages and disadvantages were meticulously documented. Some of these strategies have limits in certain scenarios, and our work will assist scientists in the fields of machine learning in selecting the most appropriate algorithms to tackle their specific problem in WSD. When comparing the works that have been used and indeed the procedures that were employed, it is possible to notice the distinctiveness of the piece of work that was created. As a result of this research, it was observed that (i) size of the dataset has an considerable impact on algorithm's performance, (ii) some methodologies provide high performance accuracy for one language where as it gives low performance for some other, (iii) a few of these methodologies can be run quickly but with a limitation on accuracy, and (iv) the large number among those methodologies have been implemented successfully for a wide range of different languages.