Sebaweh at AraGenEval Shared Task: BERENSE - BERt based ENSEmbler for Arabic Authorship Identification
Year: 2025
Type: conference-paper
Abstract: Authorship Identification for Arabic texts is challenging due to the language's dialectal diversity and the wide stylistic variation across genres, cultures, and historical periods.It has critical applications in copyright enforcement, forensic linguistics, and literary analysis.Recognizing its importance, we addressed this challenge using the AraGenEval 2025 shared task dataset, which contains works by writers from diverse backgrounds and time periods.We conducted extensive experiments with multiple architectures and proposed an ensemble model that combines the strengths of four fine-tuned transformer-based models.We applied data augmentation to enrich the dataset and class weighting to handle class imbalance during training.Our system achieved a Macro-F1 score of 90%, representing a 15% improvement over our baseline, and ranked 1st in the competition.
Authors: Muhammad Helmy, Batool Najeh Balah, Ahmed Mohamed Sallam, Ammar Sherif
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037834