Thesis Titan at CheckThat! 2023: Language-Specific Fine-tuning of mDeBERTaV3 for Subjectivity Detection

Folkert Atze Leistra, Tommaso Caselli

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

2 Citations (Scopus)
79 Downloads (Pure)

Abstract

The detection of subjectivity in natural language plays a crucial role in various applications, such as sentiment analysis, fake news detection, and fact-checking systems. However, effectively and accurately detecting subjectivity across different languages presents substantial challenges due to linguistic variations and cultural nuances. This paper describes the system we developed for 2023 CheckThat! Lab Task 2 on subjectivity detection using a multilingual model, mDeBERTaV3-base. In particular, we use a common multilingual dataset to fine-tune multiple mDeBERTaV3-base models using language specific development data to specialize the systems towards a target language and reduce the impact of the class imbalance in the training data. In this way, we managed to rank first in German, Italian and Turkish, second in Arabic and over a Multilingual dataset, and third in English.

Original languageEnglish
Title of host publicationWorking Notes of the Conference and Labs of the Evaluation Forum (CLEF 2023)
EditorsMohammad Aliannejadi, Guglielmo Faggioli, Nicola Ferro, Michalis Vlachos
PublisherCEUR Workshop Proceedings (CEUR-WS.org)
Pages351-359
Number of pages9
Publication statusPublished - 2023
Event24th Working Notes of the Conference and Labs of the Evaluation Forum, CLEF-WN 2023 - Thessaloniki, Greece
Duration: 18-Sept-202321-Sept-2023

Publication series

NameCEUR Workshop Proceedings
Volume3497
ISSN (Print)1613-0073

Conference

Conference24th Working Notes of the Conference and Labs of the Evaluation Forum, CLEF-WN 2023
Country/TerritoryGreece
CityThessaloniki
Period18/09/202321/09/2023

Keywords

  • mDeBERTaV3
  • Multilinguality
  • Sentence semantic
  • Subjectivity detection

Fingerprint

Dive into the research topics of 'Thesis Titan at CheckThat! 2023: Language-Specific Fine-tuning of mDeBERTaV3 for Subjectivity Detection'. Together they form a unique fingerprint.

Cite this