Babelfy-Based Extraction of Collocations from Turkish Hotel Reviews
Ekin Ekıncı, Sevinç İlhan Omurca · 2018
In Natural Language Processing (NLP) tasks lexical and semantic resources such as WordNet, Open Multilingual WordNet, Wikipedia, Wiktionary, OmegaWiki, BabelNet, Babelfy and so on are played an important role in terms of knowledge extraction. This study presents usage of Babelfy that, given a Turkish hotel review dataset, which contains 1517 reviews, returns all collocations the dataset contains. When the extracted collocations are evaluated according to human annotated collocations it has been seen that from 425 collocations Babelfy extract 216 collocations with 0.67 F-measure. 195 of the extracted collocations are labeled as restricted expression, 2 of the them are labeled as named-entity and 19 of them are labeled as light verbs.