Abstract:With the rise in the amount of textual data over the internet, the demand for summarizing it in a short, readable, easy-to-understand form has increased. Much of the research is being carried out to improve the efficiency of these text summarization systems. In the past, extractive text summarization was mainly carried out through human-crafted features which were unable to learn the semantic information from the text. Therefore, in an attempt to improve the quality of summary, we have designed a neural network-based completely data-driven model for extractive single-document summarization of text which we have termed as WL-AttenSumm. Our proposed model implements a Word-level Attention mechanism that focuses more on the important parts in the input sequence so relevant semantic features are captured at the word-level that helps in selecting significant sentences for the summary. Another advantage of this model is that it can extract syntactic and semantic relationships from the text by using a Convolutional Bi-GRU (Bi-directional Gated Recurrent Unit) network. We have trained our proposed model on the combined CNN/Daily Mail corpus and evaluated on the Daily Mail, combined CNN/Daily Mail, and DUC 2002 test dataset for single document summarization and obtained better results as compared to the state-of-the-art baseline approaches in terms of ROUGE metrics. For the summary length limited to 75 words, our attention-based approach generates ROUGE recall scores for R-1, R-2, R-L measures as 32.8%, 11.0%, 27.5% with Daily Mail corpus and 55.9%, 24.8%, 53.9% with DUC 2002 dataset, respectively. Experiments performed with the joint CNN/Daily dataset yield full-length ROUGE F1 scores as 42.9%, 19.7%, 39.3%. Therefore, our deep learning-based summarization framework achieves competitive performance.

Searching for Effective Neural Extractive Summarization: What Works and What's Next

Automatic Document Summarization Via Deep Neural Networks

Abstractive Summarization Improved by WordNet-based Extractive Sentences

Extractive Summarization As Text Matching

Neural Abstractive Summarization: A Brief Survey

RefSum: Refactoring Neural Summarization.

Deep learning-based extractive text summarization with word-level attention mechanism

Neural Document Summarization by Jointly Learning to Score and Select Sentences

Neural Summarization by Extracting Sentences and Words

An Editorial Network for Enhanced Document Summarization

Hie-Transformer: A Hierarchical Hybrid Transformer For Abstractive Article Summarization

A Unified Summarization Model with Semantic Guide and Keyword Coverage Mechanism

Research on Abstractive Automatic Summarization Technology Based on Deep Learning

A New Method for Extractive Text Summarization Using Neural Networks

Extractive Summarization of Documents by Combining Semantic Content and Non-Structured Features

Surveying the Landscape of Text Summarization with Deep Learning: A Comprehensive Review

DCDSum: An interpretable extractive summarization framework based on contrastive learning method

Jointly Extracting and Compressing Documents with Summary State Representations

A Survey on Neural Network-Based Summarization Methods

Neural Abstractive Text Summarization and Fake News Detection

An Abstractive Summarizer Based on Improved Pointer-Generator Network