Enhanced extractive text summarization framework for low-resourced Urdu language.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• The digital age has led to a massive increase in text data, necessitating efficient information extraction.
• Automatic text summarization aims to condense text while preserving meaning.
• While advanced summarization models exist for high-resource languages, Urdu lacks well-established methods.

Purpose of the Study:

• To propose improved supervised and unsupervised extractive text summarization models for the Urdu language.
• To address the gap in text summarization capabilities for low-resource languages.

Main Methods:

• Creation of a large-scale dataset with Urdu documents and human-annotated summaries.
• Supervised approach: Sentence feature extraction (15 features), feature reduction, and application of machine/deep learning models.
• Unsupervised approach: Utilization of four models, integrating top features from the supervised method.

Main Results:

• The proposed supervised model achieved approximately 7% improvement in ROUGE scores.
• The proposed unsupervised model achieved approximately 12% improvement in ROUGE scores.
• Both approaches demonstrated superior performance compared to existing Urdu text summarization techniques.

Conclusions:

• The developed supervised and unsupervised models offer significant advancements for Urdu text summarization.
• These models provide effective solutions for information extraction in low-resource languages.
• The research contributes to the field of natural language processing for under-resourced languages.