Abstractive Sentence Summarization with Encoder-Convolutional Neural Networks

Abstract

Summarization is the task of condensing a piece of text to produce a short version while preserving important elements and the meaning of content There have two main methods to summarize the text such as extractive summarization and abstractive summarization. Abstractive Sentence Summarization generates a shorter version of a set of documents while attempting to preserve its meaning. In this work, we introduce an architecture called the pointer-gen E-Conv (PGEC) whose conditioning is the combination between pointer-generator and a novel convolutional network with a weight normalization. Our model gains a 32.28 ROUGE-1 score on the Gigaword test set and a 27.13 ROUGE-1 score on the DUC 2004 dataset These results have shown that PGEC outperforms the recently proposed methods on both datasets.