Extractive - Abstractive Summarization Using Transformers: A Hybrid Approach

Abstract

Automatic text summarization (ATS) minimizes a lengthy text document into a condensed version by extracting the key points and main themes. Text summarising techniques can assist researchers in acquiring crucial information from various articles with less time and hassle. Several ATS systems have previously explored this area, and numerous text summarisation algorithms have been developed to extract essential details from textual sources and present them concisely. Unfortunately, many of these methods do not retain the text content's semantic elements and hidden meanings. Even though a substantial number of researchers have remedied these restrictions, they continue to provide a significant challenge for developing an effective article summarizer. This paper focuses on developing a framework for hybrid text summarization utilizing TF-IDF and Transformers to summarise research publications. Using the Recall-Oriented Understudy for Gisting Evaluation (ROUGE) metrics, the effectiveness of the proposed hybrid summarizer for 10 research articles has been compared to the gold standard summary. Results demonstrate the similarity between the automated and human-generated summaries of the input article.