Abstractive Text Summarisation using Recurrent Neural Networks at the Paragraph Level

Abstract

Continuous production of information has been facilitated by the easy access to new technology. This has made it difficult for many users to find relevant information, which are sometimes buried deeply inside mass-produced content. Without the development of new tools and technology to make this data more accessible, it potential remain unexploited. Abstractive text summarisation aims to extract the key points of the document. Because text generating techniques are still in their early stage, it has received little attention in the past. Recently, the application of recurrent neural network models has made significant progress in abstractive sentence summarisation. Despite the improvement in results, these models still tend to produce grammatical errors. Unfortunately, attempts in abstractive document summarisation are still in their early phases, and evaluation outcomes on benchmark datasets are noticeably inferior to human summarisation. In this study we propose a data-driven for abstractive document. Each word generated in the summary use an attention-based technique depending on the input paragraph. According to experimental findings, our model generates higher-quality summaries, achieving ROUGE-1 score of 44.44, ROUGE-2 score of 22.50, and ROUGEL score of 45.15 on the document understanding conference 2004 datasets.

https://doi.org/10.59200/ICARTI.2023.020