Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Dhati+: Fine-tuned Large Language Models for Arabic Subjectivity Evaluation

Created by
  • Haebom

Author

Slimane Bellaouar, Attia Nehar, Soumia Souffi, Mounia Bouameur

Outline

This paper presents a novel approach for subjectivity analysis in Arabic. While Arabic is linguistically rich and morphologically complex, the lack of large-scale annotated data hinders the development of accurate tools. This study leverages existing Arabic datasets (ASTD, LABR, HARD, SANAD) to build a comprehensive dataset, AraDhati+, and fine-tunes state-of-the-art Arabic language models (XLM-RoBERTa, AraBERT, ArabianGPT) to perform subjectivity classification. By additionally utilizing an ensemble decision-making approach, we achieved a high accuracy of 97.79%, demonstrating that this approach is effective in addressing resource constraints in Arabic processing.

Takeaways, Limitations

•
Takeaways:
◦
A novel approach for analyzing Arabic subjectivity and the AraDhati+ dataset are presented.
◦
Achieving high accuracy (97.79%) by leveraging a state-of-the-art Arabic language model.
◦
Contributing to solving the resource shortage problem in the field of Arabic natural language processing.
◦
Presenting the possibility of performance improvement through ensemble techniques.
•
Limitations:
◦
Lack of detailed description of the composition and quality of the AraDhati+ dataset.
◦
Lack of detailed description of the characteristics of the Arabic language model used and the reasons for its choice.
◦
Lack of comparative analysis with other subjectivity analysis methodologies.
◦
Lack of review of the dataset for bias and generalizability.
◦
Lack of performance evaluation in real-world applications.
👍
Made with Slashpage