CUET-NLP_Team_SS306 at AraGenEval Shared Task: A Transformer-based Framework for Detecting AI-Generated Arabic Text
Year: 2025
Type: conference-paper
Abstract: With the rapid emergence of large language models (LLMs), AI-generated content has increased, presenting new opportunities and significant risks.Detecting such content is crucial, yet while research in high-resource languages like English has advanced, work in lowresource languages, such as Arabic, remains limited.To help fill this gap, the AraGenEval 2025 workshop organized a shared task on AIgenerated Text Detection in Arabic.We participated in Task 3, where we evaluated several transformer-based models, including AraBERT, RoBERTa, AraRoBERTa, mBERT, and mar-BERT, both with and without chunking of input sequences during training.The experimental results show that applying chunking prior to training improves the performance of transformers.Among the evaluated models by the system testset, AraBERT with chunking achieved the highest F1 score (0.67), outperforming the others.Based on these results, our team ranked 12th in Shared Task 3.
Authors: Sowrav Nath, Shadman Saleh, Kawsar Ahmed, Mohammed Moshiul Hoque
Institution: Chittagong University of Engineering & Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037840