Q357 : A Multi-Level Contrastive Learning frxamework for Cross-Lingual Named Entity Recognition with a Focus on Persian Language
Thesis > Central Library of Shahrood University > Computer Engineering > MSc > 2026
Authors:
Abstarct: Abstract
Cross-lingual named entity recognition is a fundamental task in natural language processing that plays a crucial role in information extraction, knowledge retrieval, and multilingual text analysis. Despite significant advances in multilingual language models, their performance on low-resource languages such as Persian remains limited due to the scarcity of labeled data, semantic misalignment across languages, and insufficient modeling of structural relationships between words. Previous approaches have mainly relied on sequence labeling and direct knowledge transfer from high-resource languages, which often leads to degraded performance, especially in Persian texts. In this study, we propose a frxamework baxsed on multi-laxyer contrastive learning and structural word relationship modeling to address these challenges. The proposed method leverages hybrid bilingual data and self-supervised learning to reduce the gap between English and Persian representations. Additionally, the use of unlabeled Persian data improves the model’s generalization capability. Experimental results demonstrate that the proposed frxamework outperforms baxseline methods, improving the F1-score in Persian from approximately 0.29 to 0.47. Furthermore, for specific entity classes such as PER, the F1-score reaches approximately 0.77. These improvements indicate the effectiveness of the proposed approach in enhancing accuracy and stability in cross-lingual named entity recognition tasks.
Keywords:
#Keywords: Named Entity Recognition #Contrastive Learning #Multilingual Language Models #Persian NLP #Self-Supervised Learning Keeping place: Central Library of Shahrood University
Visitor:
Visitor: