Transfer Learning Across Arabic Dialects for Offensive Language Detection

The Arabic language is spoken by a wide range of countries across Asia, however, it is a low-resource language that has a minimal number of linguistic resources. Moreover, the large spread of Arabic speakers spans several countries and cultures, which creates a complex variation in its dialectal for...

Full description

Saved in:
Bibliographic Details
Published in2022 International Conference on Asian Language Processing (IALP) pp. 196 - 205
Main Authors Husain, Fatemah, Uzuner, Ozlem
Format Conference Proceeding
LanguageEnglish
Published IEEE 27.10.2022
Subjects
Arabic dialects
BERT customization
BERT model
Bit error rate
Data models
Libraries
Linguistics
Offensive language detection
Task analysis
text classification
Training
Transfer learning
Online AccessGet full text
DOI10.1109/IALP57159.2022.9961263

Cover

More Information
Summary:The Arabic language is spoken by a wide range of countries across Asia, however, it is a low-resource language that has a minimal number of linguistic resources. Moreover, the large spread of Arabic speakers spans several countries and cultures, which creates a complex variation in its dialectal form. This variation makes it very challenging to analyze online Arabic content, particularly for offensive language detection. We propose a transfer learning approach for dialectal Arabic offensives language detection based on the BERT model. The results demonstrate the effectiveness of the proposed system in improving the performance of some Arabic dialects, such as the Tunisian and the Egyptian.
DOI:10.1109/IALP57159.2022.9961263