Automatic Text Summarization Using Customizable Fuzzy Features and Attention on the Context and Vocabulary

Abstract

Nowadays, as the users of the internet are increasingly ensnared by information overload, the significance of automatic text summarization becomes more and more evident. Thus far, the performance of the proposed approaches to automatically summarize complex articles, e.g. news articles and technical documents, have been far away from application levels to produce humanlike summaries, especially when the complexity of such articles are increased by discussing multiple topics. In this research, we present a novel automatic, yet more applicable, approach for text summarization using customizable fuzzy features and neural sequence-to-sequence model with attention mechanism on the word distribution both in the context and vocabulary. We assess the performance of our model on CNN/Daily Mail summarization task, through multiple devised evalution schemes comparing its applicability and summaries' quality with the state-of-the-art models.