Abstractive Summarization Using Neural Networks with Attention Mechanisms

Surendrakumar, Kruthika (2024) Abstractive Summarization Using Neural Networks with Attention Mechanisms. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (7MB) | Preview

Abstract

In the present world where a huge quantity of information is available, text summarization especially news articles text summarization has gained higher importance. Previous approaches to performing traditional extractive summarization appear to work well but fails in terms of providing precise and meaningful summaries. Previous methods and researched tended to suffer from low performance of generating coherent and informative summaries that would capture content of news articles. In contrast to the earlier works done in this area and to overcome the shortcomings, this study presents an improved abstractive summarization model for news articles using advanced sequence to sequence model framework. This study explores the methods of news text summarization with sequence-to-sequence models. Using the CNN/DailyMail dataset that offers more than 300,000 articles, the following piece of work focuses on the performance of basic and extended models to create ‘compressed’ and ‘cohesive’ summaries. The study evaluates two baseline models: LSTM and Bidirectional LSTM both Random Search regimen exclusion, and No-Attention paradigm. It then generalises this assessment to models with attention and learnt pretrained GloVe word vectors, namely Sequence to Sequence LSTM with attention and embedding, and BiLSTM with attention and embedding. The overall objective is therefore to evaluate the impact of these enhancements to the quality of summaries via ROUGE scores. It brings out an all round collection of data, cleaning and preprocessing of text data before proceeding to the final stages of model building. The performance is highly guarded and measured through the ROUGE metrics that allow the assessment of the quality of the generated summaries between the models.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Shahid, Abdul UNSPECIFIED
Uncontrolled Keywords:	News text summarization; Sequence-to-sequence models; BiLSTM; Attention Mechanism; Glove Embeddings; Abstractive Summarization
Subjects:	N Fine Arts > NE Print media Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science P Language and Literature > P Philology. Linguistics > Computational linguistics. Natural language processing
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	05 Sep 2025 11:21
Last Modified:	05 Sep 2025 11:21
URI:	https://norma.ncirl.ie/id/eprint/8821

Actions (login required)

View Item