Novel Attention Deep Learning Model for Cross Language Abstractive Text Summarization

Abstract

In the age of information, the huge amount of data that is produced every day will remain unuseful for humans unless we make it available with new tools and technologies. Abstractive Text Summarization tries to get the most essential content of a text corpus and compress it into a shorter text while keeping its meaning and maintaining its semantic and grammatical correctness. Neural architectures are becoming dominant in Abstractive Text Summarization. The use of deep learning architectures in natural language processing entered a new area after the appearance of seq2sep models in recent decades. These models are basically founded on a couple of recurrent neural networks that connect the input and output data in an encoder-decoder architecture. Better results were possible by adding the Attention Mechanism to the RNN layers. Many studies have shown how competitive these new architectures are in machine translation. The recently proposed model, which is called Abstractive Text Summarization (ATS), uses only the attention mechanism. It has surpassed the previous state-of-the-art models in translation tasks. In this work, we want to experiment with this new model for abstractive text summarization. The evaluation results on the Amazon fine food reviews dataset show that our method better captures the attention pattern and achieves The Bleu scored equivalent (0,50). The result is a very high-quality summary. Abstractive Cross-Language Summarization CLA (translation and summarization) is the task of generating a summary in a target language from a collection of documents in a different source language. The result of the blue score was very good and satisfying (0.50).