Due to the exponential growth of online textual data, and rapid expansion of internet usage, tasks such as document management, text classification, and information retrieval have been significantly posing a challenge due to its huge size. A key tool for addressing this issue is Automatic Text Summarization (ATS). It is one of the most challenging tasks since there is no precise computational method for evaluating accuracy of a summary.[1] The main function of ATS is to automatically create a concise and apprehensive summary by extracting the main ideas from the source text. Models for abstractive summarization based on deep learning (DL) have recently been created to better balance and improve these two elements. The field of DL-based text summarization currently lacks a thorough literature review, nevertheless. This work offers researchers a thorough analysis of DL-based text summarization in order to close this gap. In this paper, many prevalent frameworks for text summarization are listed. We also propose Summa - an Abstractive text summarizer which uses Long Short Term Memory(LSTM) based Sequence to Sequence Model. Accuracy matrix or F1 score are not recommended, traditional evaluation measures for text summarisation are not that much relevant so, a novel measure, Bi-Lingual Evaluation Understudy(BLEU)[2] measure is used. The study reported the BLEU score of 0.71, with attention mechanism and 0.3, without Attention Mechanism, which is considered a good score.

Extractive Abstractive Long Short Term Memory( Stacked LSTM) Attention BLEU.