Bangla-ExtraSum: Comparative Analysis of Different Methods in Automated Extractive Bengali Text Summarization

Text summarization is one of the most exciting fields for modern researchers. Through summarization, the reader can gain a comprehensive understanding of the larger content. Summarization techniques can be applied in every content area. The demand for summarization in news portals is growing day by day. Though several approaches for summarizing English texts have been developed, only a few notable works have been done in Bangla. In addition, there are just a handful of Bangla extractive summarization models and datasets available. In this paper, we introduce different approaches to text summarization that can capture the main context of the text using semantic and contextual relations between sentences. There is also a visual representation of the comprehensive analysis of these approaches and their results. The necessity of semantic relations between sentences while doing summarization is shown here and also demonstrated in transformer model and word2vec model assistance in summarization methodologies. 500 articles are proposed with two human-generated summaries for each article in this research. Finally, the performances of five different types of models in two different datasets are shown. Our designed model gets a 0.68 f-score in the existing dataset and a 0.63 f-score in our dataset.