NLP_wizard at AraGenEval shared task: Embedding-Based Classification for AI Detection and Authorship Attribution
Year: 2025
Type: conference-paper
Abstract: This paper presents a lightweight system for the AraGenEval shared task, addressing AI-generated text detection and authorship identification in Arabic.Using pretrained xlm-roberta-large embeddings with mean pooling and [CLS] token strategies, combined with classical classifiers (RidgeClassifierCV and LinearSVC), our approach achieved F1scores of 0.7400 and 0.8130 on the ARATECT and authorship datasets, respectively.Mean pooling outperformed [CLS] by 3%, demonstrating efficiency and robustness for limited Arabic data while capturing stylistic nuances critical for both tasks.
Author: Mena Hany
Institution: King Fahd University of Petroleum and Minerals
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037670