Enhanced extractive text summarization framework for low-resourced Urdu language

Abstract

This era has witnessed an enormous increase in textual corpus available in digital form. Therefore, an intelligent mechanism is required to extract the essential information. This task is performed using an automatic text summarization that converts the text into a shorter form while the semantics are preserved. The popular languages of the world such as English, Chinese etc. have well-developed text summarization models. However, for low-resourced languages such as Urdu, well-established methods are missing. This research work proposes improved supervised and unsupervised extractive text summarization models. A large-scale dataset containing text documents and their human-annotated extractive summaries has been created. In the supervised approach, fifteen features are extracted against each sentence in the text. Further, to reduce the computational complexity, feature reduction is performed. Multiple machine learning and deep learning models are employed, including the proposed model. In an unsupervised approach, four different models are utilized. Further, the high results-producing model is incorporated with the top three features that significantly contributed to the supervised approach. The results are evaluated using ROUGE scores. Experimental results demonstrate that the proposed supervised and unsupervised approaches outperform existing Urdu text summarization methods, achieving approximately 7% and 12% improvements in ROUGE scores, respectively.