Representation of Words in Natural Language Processing: ? Survey
Тип публікації :
Стаття
Дата випуску :
2019
Автор(и) :
Losieva, Y. A.
Мова основного тексту :
Англійська
eKNUTSHIR URL :
Випуск :
2
ISSN :
1812-5409
Початкова сторінка :
82
Кінцева сторінка :
87
Цитування :
[APA 7] Losieva, Y. A. (2019). Representation of Words in Natural Language Processing: ? Survey. Bulletin of Taras Shevchenko National University of Kyiv. Physics and Mathematics, (2), 82–87. https://doi.org/10.17721/1812-5409.2019/2.10
[ДСТУ] Losieva Y. A. Representation of Words in Natural Language Processing: ? Survey. Bulletin of Taras Shevchenko National University of Kyiv. Physics and Mathematics. 2019. no. 2. P. 82—87. DOI: 10.17721/1812-5409.2019/2.10 (date of access: 25.07.2026).
The article is devoted to research to the state-of-art vector representation of words in natural language processing. Three main types of vector representation of a word are described, namely: static word embeddings, use of deep neural networks for word representation and dynamic) word embeddings based on the context of the text. This is a very actual and much-demanded area in natural language processing, computational linguistics and artificial intelligence at all. Proposed to consider several different models for vector representation of the word (or word embeddings), from the simplest (as a representation of text that describes the occurrence of words within a document or learning the relationship between a pair of words) to the multilayered neural networks and deep bidirectional transformers for language understanding, are described chronologically in relation to the appearance of models. Improvements regarding previous models are described, both the advantages and disadvantages of the presented models and in which cases or tasks it is better to use one or another model.Key words: artificial intelligence, natural language processing, computational linguistics, word embeddings.Pages of the article in the issue: 82 - 87Language of the article: English
Файл(и) :![Ескіз]()
Вантажиться...
Формат :
Adobe PDF
Розмір :
306.75 KB
Контрольна сума :
(MD5):eba33408021fe0401ef3c73c134af70a
Якщо не вказано інше, ця робота розповсюджується на умовах ліцензії Creative Commons Attribution 4.0 International