Рад у зборнику

објављена

енглески

Branislava Šandrih Todorović, undefined undefined, Cvetana Krstev, Ranka Stanković, Milica Ikonić Nešić, undefined undefined, undefined undefined, undefined undefined

Serbian NER&Beyond: The Archaic and the Modern Intertwinned

Proceedings of the Conference Recent Advances in Natural Language Processing - Deep Learning for Natural Language Processing Methods and Applications

Galia Angelova, Maria Kunilovskaya, Ruslan Mitkov, Ivelina Nikolova-Koleva

INCOMA Ltd. Shoumen, BULGARIA

2021

U ovom radu predstavljamo srpski književni korpus koji se razvija pod okriljem COST Akcije „Distant Reading for European Literary History” CA16204. Koristeći ovaj korpus romana napisanih pre više od jednog veka, razvili smo i učinili javno dostupnim Sistem za prepoznavanje imenovanih entiteta (NER) obučen da prepozna 7 različitih tipova imenovanih entiteta, sa konvolucionom neuronskom mrežom (CNN), koja ima F1 rezultat od ≈91% na test skupu podataka. Ovaj model je dalje ocenjen na posebnom skupu podataka za evaluaciju. Završavamo poređenje razvijenog modela sa postojećim, nakon čega sledi diskusija o prednostima i nedostacima oba modela.

In this work, we present a Serbian literary corpus that is being developed under the umbrella of the “Distant Reading for European Literary History” COST Action CA16204. Using this corpus of novels written more than a century ago, we have developed and made publicly available a Named Entity Recognizer (NER) trained to recognize 7 different named entity types, with a Convolutional Neural Network (CNN) architecture, having F1 score of ≈91% on the test dataset. This model has been further assessed on a separate evaluation dataset. We wrap up with comparison of the developed model with the existing one, followed by a discussion of pros and cons of the both models.

1252

1260

10.26615/978-954-452-072-4_141

978-954-452-072-4

http://dx.doi.org/10.26615/978-954-452-072-4_141

М30

М33

Отворени приступ

Creative Commons – Attribution-Share Alike 4.0 International

.pdf