# Automated News Summarization Using Transformers

**Authors:** Anushka Gupta, Diksha Chugh,  Anjum, Rahul Katarya  
**arXiv:** [2108.01064](https://arxiv.org/abs/2108.01064)  
**Date:** 2021-04-23  
**Summary page:** https://www.summarizepaper.com/en/paper/2108.01064v1/

## Key points

- The amount of text data available online is growing rapidly, emphasizing the need for automated summarization in modern recommender and text classification systems.
- Two main methods of generating summaries are extractive summarization, which selects relevant sentences from the original document, and abstractive summarization, which interprets the text to generate a summary.
- A study by Anushka Gupta, Diksha Chugh, Anjum, and Rahul Katarya from Delhi Technological University compares extractive and abstractive methods for text summarization using the BBC news dataset.
- Automating summarization processes can save time, reduce manual efforts, optimize storage space with shorter texts, and play a vital role in text mining and data analysis.
- Extractive summarization involves selecting important phrases or sentences based on computed scores, while abstractive summarization predicts a summary by paraphrasing sections of the original document.
- The research focuses on abstractive summarization due to its complexity in simulating human perception for developing accurate and fluent summaries.
- The study aims to enhance understanding of transformer-based pre-trained models for text summarization using real-world datasets like BBC news articles.

## Summary

The amount of text data available online is growing rapidly, making  a crucial tool for modern recommender and text classification systems. Manually creating concise summaries of lengthy articles is time-consuming and tedious, highlighting the need for automated summarization to train machine learning models efficiently. Two main methods of generating summaries are , which selects relevant sentences from the original document, and , which interprets the text to generate a summary. In this paper by Anushka Gupta, Diksha Chugh, Anjum, and Rahul Katarya from Delhi Technological University in New Delhi, India, a comprehensive comparison of  for text summarization is presented. The study utilizes the BBC news dataset for analysis and comparison purposes, using human-generated summaries as benchmarks. The introduction emphasizes the importance of news summarization in creating concise summaries without losing essential information. Automating summarization processes can reduce manual efforts and reading time while optimizing storage space with shorter texts. Accurate summaries play a vital role in text mining and data analysis. Summarization techniques are classified into  and . Extractive summarization involves selecting important phrases or sentences from the text based on computed scores. On the other hand, abstractive summarization interprets the text to predict a summary by paraphrasing sections of the original document. The focus of this work is on  due to its complexity in simulating human perception for developing accurate and fluent summaries. This research aims to enhance understanding of transformer-based pre-trained models for text summarization through an in-depth comparison using real-world data sets like BBC news articles. Overall, this study contributes to advancing natural language processing and deep learning techniques in the field of text summarization with transformers as key components for improving efficiency and accuracy in generating automated summaries.