Open AccessTransformer-based Sarcasm Detection in English and Slovene Language
Author(s) -
Matic Rašl,
Mitja Žalik,
Vid Keršič
Publication year - 2021
Language(s) - English
Resource type - Conference proceedings
DOI - 10.18690/978-961-286-516-0.10
Subject(s) - sarcasm , transformer , computer science , natural language processing , artificial intelligence , artificial neural network , linguistics , engineering , irony , philosophy , voltage , electrical engineering
Sarcasm detection is an important problem in the field of natural language processing. In this pa-per, we compare performances of the three neural networks for sarcasm detection on English and Slovene datasets. Each network is based on a di˙erent transformer model: RoBERTa, Distil-Bert, and DistilBert – multilingual. In addition to the existing Twitter-based English dataset, we also created the Slovene dataset using the same approach. An F1 score of 0.72 and 0.88 was achieved in the English and Slovene dataset, re-spectively.