ARC-NLP at CheckThat! 2022: Contradiction for Harmful Tweet Detection

Date

2022-01-01

Author

Toraman, Çağrı
Ozcelik, Oguzhan
Şahinuç, Furkan
Sahin, Umitcan

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

19
views

0
downloads

Cite This

The target task of our team in CLEF2022 CheckThat! Lab challenge is Task-1C, harmful tweet detection. We propose a novel approach, called ARC-NLP-contra, which is a contradiction check approach by using the idea that harmful tweets contradict with the real-life facts in the scope of COVID-19 pandemic. Besides, we propose and examine two other models. The first model, called ARC-NLP-hc, is a traditional approach that utilizes hand-crafted tweet and user features. The second model, called ARC-NLP-pretrain, pretrains a Transformer-based language model by using COVID-related Turkish tweets. We compare the performances of these three models, and submit the highest performing model in the preliminary experiments to the challenge. We make submissions for Task-1A, 1B, 1C in Turkish and Task-1C in English. We have the winning solution for Task-1C, harmful tweet detection in Turkish, using ARC-NLP-contra that is our contradiction check approach.

Subject Keywords

Claim Detection, Contradiction, COVID-19, Harmful Tweet Detection, Language Model, Tweet, Worthiness Checking

URI

https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85136961444&origin=inward
https://hdl.handle.net/11511/109704

Collections

Department of Computer Engineering, Conference / Seminar