A deep learning approach for Named Entity Recognition in Urdu language.

Area of Science:

• Natural Language Processing
• Machine Learning
• Computational Linguistics

Background:

• Named Entity Recognition (NER) is crucial for information extraction but remains under-researched for Urdu due to linguistic complexities.
• Existing Urdu NER models often rely on word embeddings for feature extraction, with varying success.

Purpose of the Study:

• To propose a robust deep learning approach for Urdu Named Entity Recognition (NER).
• To enhance feature extraction by leveraging FastText and Floret word embeddings for contextual information.
• To evaluate the proposed model against state-of-the-art methods on benchmark Urdu datasets.

Main Methods:

• Utilized pre-trained FastText and Floret word embeddings to generate feature vectors for four Urdu datasets.
• Trained various deep learning models, including Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU) with Conditional Random Field (CRF).
• Implemented combinations of BiLSTM+GRU with Floret embeddings as the primary model architecture.

Main Results:

• The proposed deep learning approach significantly outperformed existing state-of-the-art Urdu NER methods.
• Achieved a maximum F-score of 0.98 using the BiLSTM+GRU architecture with Floret embeddings.
• Demonstrated robustness with a low classification error rate, ranging from 1.24% to 3.63% across datasets.

Conclusions:

• The integration of FastText and Floret embeddings with advanced deep learning architectures (BiLSTM+GRU) offers a highly effective solution for Urdu NER.
• The proposed method represents a significant advancement in Urdu Natural Language Processing, addressing the challenges posed by the language's morphology.
• The achieved performance indicates the potential for broader applications of this approach in Urdu text analysis and information retrieval.