Abstractive Bengali Text Summarization Using Transformer-based Learning

Text summarization serves the purpose of condensing extensive text documents into concise sentences that capture the core information. Given the exponential growth in textual data due to web-based applications, there is a pressing need for automated systems to streamline the summarization process, saving both time and resources. In response to this demand, this paper introduces an automated abstractive text summarization framework for Bengali, leveraging transformer-based models. The research evaluates the performance of five different transformer-based approaches using available datasets. The results demonstrate that the Bangla-T5 model, fine-tuned on a benchmark dataset, outperforms other methods, achieving the highest ROUGE-2 score of 13.83 on the test dataset.