Zero and Few-shot Hate Speech Detection in Social Media Messages Related to Earthquake Disaster

Date

2023-01-01

Author

Sahin, Umitcan
Kucukkaya, Izzet Emre
Ozcelik, Oguzhan
Toraman, Çağrı

Metadata

Show full item record

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

Item Usage Stats

17
views

0
downloads

Cite This

Microblog platforms provide active communication channels during crisis situations such as earthquakes. Data from these platforms plays an important role in alleviating such crises. In times of crisis, it is important to develop quick and reliable methods. In this study, we perform hate speech detection in earthquake-related social media messages using zero-shot and few-shot methods that work with minimal labeled data. We collect Turkish messages shared on Twitter from the 2023 Turkey-Syria earthquake and label them according to their hate speech content. First, we evaluate the zero-shot performance of the proposed models that are trained on a different hate speech dataset on the newly created earthquake dataset. Then, using the newly labeled data, we fine-tune the models with few-shot learning, and evaluate their performance on the same dataset. The results show that transformer-based language models perform better than traditional models when there is not enough labeled data, and the few-shot method outperforms the zero-shot method. Additionally, we examine the effect of varying the size of the training data used for few-shot learning.

URI

https://hdl.handle.net/11511/109696

Collections

Department of Computer Engineering, Conference / Seminar