ENHANCED TEXT SUMMARIZATION WITH GLOBAL AND LOCAL CONTEXT

Fuente: Zenodo
Enregistré dans:
Détails bibliographiques
Auteur principal: Leela Prasad
Format: Recurso digital
Langue:anglais
Publié: Zenodo 2025
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866901204977582080
author Leela Prasad
author_facet Leela Prasad
contents <p>In recent years, encoder–decoder architectures have become fundamental to abstractive text summarization, <br>enabling models to process variable-length documents efficiently. Among these, transformer-based systems <br>leverage attention mechanisms that connect distant parts of a text, strengthening their contextual comprehension. <br>However, such global context often overlooks hierarchical detail, requiring additional modelling strategies to <br>capture both broader and finer document structures. <br>This study introduces an enhanced summarization framework that integrates prior knowledge to balance global <br>and local understanding. The proposed design employs two complementary encoding mechanisms: global <br>information-aware encoding, which generates a holistic representation of the document to guide summary <br>generation, and local information-aware encoding, which applies convolutional operations to capture fine-grained <br>textual features. Evaluations on the LCSTS and CSL datasets demonstrate improved ROUGE scores compared to <br>baseline and leading benchmark systems, confirming that integrating multi-level representations enhances <br>summary accuracy and informativeness.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_17559075
institution Zenodo
language eng
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle ENHANCED TEXT SUMMARIZATION WITH GLOBAL AND LOCAL CONTEXT
Leela Prasad
Abstractive Summarization, Encoder-Decoder Architecture, Transformer, Global and Local Encoding, Natural Language Processing, Machine Learning
<p>In recent years, encoder–decoder architectures have become fundamental to abstractive text summarization, <br>enabling models to process variable-length documents efficiently. Among these, transformer-based systems <br>leverage attention mechanisms that connect distant parts of a text, strengthening their contextual comprehension. <br>However, such global context often overlooks hierarchical detail, requiring additional modelling strategies to <br>capture both broader and finer document structures. <br>This study introduces an enhanced summarization framework that integrates prior knowledge to balance global <br>and local understanding. The proposed design employs two complementary encoding mechanisms: global <br>information-aware encoding, which generates a holistic representation of the document to guide summary <br>generation, and local information-aware encoding, which applies convolutional operations to capture fine-grained <br>textual features. Evaluations on the LCSTS and CSL datasets demonstrate improved ROUGE scores compared to <br>baseline and leading benchmark systems, confirming that integrating multi-level representations enhances <br>summary accuracy and informativeness.</p>
title ENHANCED TEXT SUMMARIZATION WITH GLOBAL AND LOCAL CONTEXT
topic Abstractive Summarization, Encoder-Decoder Architecture, Transformer, Global and Local Encoding, Natural Language Processing, Machine Learning
url https://doi.org/10.5281/zenodo.17559075