Word Sense Disambiguation using KeNet

Abstract

The highly studied Natural Language Processing (NLP) problem Word Sense Disambiguation (WSD) is the process of removing the ambiguities of multiple-sense words that have the same morphological structure. The first step of WSD is to list the probable meanings of the word.The next step is identify the meaning which is used in the context within the sentence. A Turkish WordNet called KeNet is used to list the word-senses. The elimination of the ambiguity was done with BERT word embeddings by comparing with the meaning via cosine similarity. The impact of the system is evaluated by both with and without adding it to a search engine of Turkish news. Each news of topmost 10 news returned for each query is manually labeled as related or not related.Results of the labeling, relatedness of the first n documents, and ordered biased precision metrics are evaluated. Positive increment on the results is shown when the WSD modul is added on the system.