Abstractive Summarization of Text using Encoder-Decoder Based Architecture

The internet keeps bringing tons and tons of information to its users on a daily basis – reading everything can consume months or years and sometimes even decades. Access to this much information has helped us in several disciplines, but at times we are overwhelmed with information and end up getting confused. To solve this to an extent possible there are summarization techniques that automatically reduce the given text to a considerable size that can be easily studied. Such large text documents are quite impossible for humans to summarize and might end up useless since the scope for the information might change during the summarization process. This is where computers come into the scene. Computers are way better in handling large amount of data than humans since it can easily do repetitive tasks with accuracy and speed. Computers summarize and give us a comparatively small notes like document to simplify our workload. Though there are several techniques widely used for summarizing documents they can be widely classified into two categories - abstractive and extractive summarization. Many projects have been published on extraction summarization, however, it cannot provide summary close to human language. We try to provide summary close to human language using abstractive text summarization method. This project uses neural network models for abstractive summarization on long texts.